diff --git "a/data/val.json" "b/data/val.json" deleted file mode 100644--- "a/data/val.json" +++ /dev/null @@ -1,20401 +0,0 @@ -{ - "dataset_info": { - "name": "val", - "total_scenes": 210, - "total_qa": 1954, - "question_type_counts": { - "target presence verification": 515, - "nearest reference identification": 454, - "fine-grained spatial description": 490, - "structured spatial prediction": 495 - }, - "dataset_qa_counts": { - "scannet200": 600, - "scannetpp": 1354 - }, - "selected_scenes": [ - { - "dataset": "scannet200", - "scene_name": "scene0006_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0024_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0025_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0034_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0040_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0042_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0047_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0048_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0070_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0083_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0084_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0086_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0089_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0095_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0097_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0100_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0101_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0109_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0117_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0121_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0136_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0140_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0156_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0164_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0166_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0188_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0191_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0196_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0207_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0211_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0212_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0213_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0220_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0222_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0225_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0226_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0233_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0252_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0255_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0262_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0264_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0274_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0278_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0288_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0304_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0310_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0327_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0331_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0333_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0336_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0342_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0347_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0349_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0352_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0353_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0358_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0369_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0370_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0404_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0409_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0430_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0434_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0438_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0446_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0476_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0477_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0508_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0509_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0518_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0522_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0532_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0535_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0538_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0547_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0563_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0564_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0566_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0571_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0580_00" - }, - { - "dataset": "scannet200", - "scene_name": "scene0582_00" - }, - { - "dataset": "scannetpp", - "scene_name": "00dd871005" - }, - { - "dataset": "scannetpp", - "scene_name": "2e74812d00" - }, - { - "dataset": "scannetpp", - "scene_name": "354e4a0a50" - }, - { - "dataset": "scannetpp", - "scene_name": "3583c4c70f" - }, - { - "dataset": "scannetpp", - "scene_name": "3597f00367" - }, - { - "dataset": "scannetpp", - "scene_name": "35f2120068" - }, - { - "dataset": "scannetpp", - "scene_name": "3799bd47b3" - }, - { - "dataset": "scannetpp", - "scene_name": "39705fbdf1" - }, - { - "dataset": "scannetpp", - "scene_name": "39f36da05b" - }, - { - "dataset": "scannetpp", - "scene_name": "3a3745a437" - }, - { - "dataset": "scannetpp", - "scene_name": "3aa115e55e" - }, - { - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3" - }, - { - "dataset": "scannetpp", - "scene_name": "3eba679830" - }, - { - "dataset": "scannetpp", - "scene_name": "3ee9d33aef" - }, - { - "dataset": "scannetpp", - "scene_name": "3f15a9266d" - }, - { - "dataset": "scannetpp", - "scene_name": "3f1e1610de" - }, - { - "dataset": "scannetpp", - "scene_name": "40b56bf310" - }, - { - "dataset": "scannetpp", - "scene_name": "413085a827" - }, - { - "dataset": "scannetpp", - "scene_name": "43cd995c51" - }, - { - "dataset": "scannetpp", - "scene_name": "4422722c49" - }, - { - "dataset": "scannetpp", - "scene_name": "4423a61d09" - }, - { - "dataset": "scannetpp", - "scene_name": "442b144761" - }, - { - "dataset": "scannetpp", - "scene_name": "469f112e38" - }, - { - "dataset": "scannetpp", - "scene_name": "47c1209b9e" - }, - { - "dataset": "scannetpp", - "scene_name": "48573f4c95" - }, - { - "dataset": "scannetpp", - "scene_name": "48701abb21" - }, - { - "dataset": "scannetpp", - "scene_name": "48da625e8c" - }, - { - "dataset": "scannetpp", - "scene_name": "49a82360aa" - }, - { - "dataset": "scannetpp", - "scene_name": "54b6127146" - }, - { - "dataset": "scannetpp", - "scene_name": "5748ce6f01" - }, - { - "dataset": "scannetpp", - "scene_name": "57d9fecae9" - }, - { - "dataset": "scannetpp", - "scene_name": "5f0fb991a7" - }, - { - "dataset": "scannetpp", - "scene_name": "610eff6509" - }, - { - "dataset": "scannetpp", - "scene_name": "612f70fe00" - }, - { - "dataset": "scannetpp", - "scene_name": "618310ed87" - }, - { - "dataset": "scannetpp", - "scene_name": "639f2c4d5a" - }, - { - "dataset": "scannetpp", - "scene_name": "64b8886d68" - }, - { - "dataset": "scannetpp", - "scene_name": "666d04a14a" - }, - { - "dataset": "scannetpp", - "scene_name": "67ab69f888" - }, - { - "dataset": "scannetpp", - "scene_name": "6855e1ac32" - }, - { - "dataset": "scannetpp", - "scene_name": "689fec23d7" - }, - { - "dataset": "scannetpp", - "scene_name": "6c75a5290e" - }, - { - "dataset": "scannetpp", - "scene_name": "6cc2231b9c" - }, - { - "dataset": "scannetpp", - "scene_name": "6e420fd87a" - }, - { - "dataset": "scannetpp", - "scene_name": "6ef2ac745a" - }, - { - "dataset": "scannetpp", - "scene_name": "70945f435a" - }, - { - "dataset": "scannetpp", - "scene_name": "73f9370962" - }, - { - "dataset": "scannetpp", - "scene_name": "778769dad8" - }, - { - "dataset": "scannetpp", - "scene_name": "7968e15e5e" - }, - { - "dataset": "scannetpp", - "scene_name": "7a629f789c" - }, - { - "dataset": "scannetpp", - "scene_name": "7b04052ad0" - }, - { - "dataset": "scannetpp", - "scene_name": "7bc286c1b6" - }, - { - "dataset": "scannetpp", - "scene_name": "7c31a42404" - }, - { - "dataset": "scannetpp", - "scene_name": "7f77abce34" - }, - { - "dataset": "scannetpp", - "scene_name": "8890d0a267" - }, - { - "dataset": "scannetpp", - "scene_name": "891865308a" - }, - { - "dataset": "scannetpp", - "scene_name": "893fb90e89" - }, - { - "dataset": "scannetpp", - "scene_name": "8b5caf3398" - }, - { - "dataset": "scannetpp", - "scene_name": "8be0cd3817" - }, - { - "dataset": "scannetpp", - "scene_name": "8fc40ba77b" - }, - { - "dataset": "scannetpp", - "scene_name": "927aacd5d1" - }, - { - "dataset": "scannetpp", - "scene_name": "96b7c379eb" - }, - { - "dataset": "scannetpp", - "scene_name": "9816c49e97" - }, - { - "dataset": "scannetpp", - "scene_name": "99a2503d63" - }, - { - "dataset": "scannetpp", - "scene_name": "99fa5c25e1" - }, - { - "dataset": "scannetpp", - "scene_name": "9e9d017ad1" - }, - { - "dataset": "scannetpp", - "scene_name": "9f139a318d" - }, - { - "dataset": "scannetpp", - "scene_name": "a1d9da703c" - }, - { - "dataset": "scannetpp", - "scene_name": "a30b666f83" - }, - { - "dataset": "scannetpp", - "scene_name": "a492fe77aa" - }, - { - "dataset": "scannetpp", - "scene_name": "a4e227f506" - }, - { - "dataset": "scannetpp", - "scene_name": "a4f10da05d" - }, - { - "dataset": "scannetpp", - "scene_name": "a7631adbe7" - }, - { - "dataset": "scannetpp", - "scene_name": "a766294249" - }, - { - "dataset": "scannetpp", - "scene_name": "a892730b61" - }, - { - "dataset": "scannetpp", - "scene_name": "ab6983ae6c" - }, - { - "dataset": "scannetpp", - "scene_name": "ad562ce09b" - }, - { - "dataset": "scannetpp", - "scene_name": "af112b8903" - }, - { - "dataset": "scannetpp", - "scene_name": "b0b004c40f" - }, - { - "dataset": "scannetpp", - "scene_name": "b2632b738a" - }, - { - "dataset": "scannetpp", - "scene_name": "b26e64c4b0" - }, - { - "dataset": "scannetpp", - "scene_name": "b37c033757" - }, - { - "dataset": "scannetpp", - "scene_name": "b8d441d736" - }, - { - "dataset": "scannetpp", - "scene_name": "bb05a0c48c" - }, - { - "dataset": "scannetpp", - "scene_name": "bc400d86e1" - }, - { - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d" - }, - { - "dataset": "scannetpp", - "scene_name": "bd7375297e" - }, - { - "dataset": "scannetpp", - "scene_name": "bd9305480d" - }, - { - "dataset": "scannetpp", - "scene_name": "c06a983e63" - }, - { - "dataset": "scannetpp", - "scene_name": "c07c8a276d" - }, - { - "dataset": "scannetpp", - "scene_name": "c0c863b72d" - }, - { - "dataset": "scannetpp", - "scene_name": "c285c82ade" - }, - { - "dataset": "scannetpp", - "scene_name": "c2d714d386" - }, - { - "dataset": "scannetpp", - "scene_name": "c413b34238" - }, - { - "dataset": "scannetpp", - "scene_name": "c465f388d1" - }, - { - "dataset": "scannetpp", - "scene_name": "c856c41c99" - }, - { - "dataset": "scannetpp", - "scene_name": "c8d099ecd8" - }, - { - "dataset": "scannetpp", - "scene_name": "ca0c580422" - }, - { - "dataset": "scannetpp", - "scene_name": "ce12db9e81" - }, - { - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f" - }, - { - "dataset": "scannetpp", - "scene_name": "d918af9c5f" - }, - { - "dataset": "scannetpp", - "scene_name": "dbd97b5bca" - }, - { - "dataset": "scannetpp", - "scene_name": "dd636ce603" - }, - { - "dataset": "scannetpp", - "scene_name": "de5881aa12" - }, - { - "dataset": "scannetpp", - "scene_name": "dfac5b38df" - }, - { - "dataset": "scannetpp", - "scene_name": "e0de253456" - }, - { - "dataset": "scannetpp", - "scene_name": "e3b3b0d0c7" - }, - { - "dataset": "scannetpp", - "scene_name": "e3ef8b690b" - }, - { - "dataset": "scannetpp", - "scene_name": "e51b753e48" - }, - { - "dataset": "scannetpp", - "scene_name": "e6372539da" - }, - { - "dataset": "scannetpp", - "scene_name": "e667e09fe6" - }, - { - "dataset": "scannetpp", - "scene_name": "e7d7e7799d" - }, - { - "dataset": "scannetpp", - "scene_name": "e8ea9b4da8" - }, - { - "dataset": "scannetpp", - "scene_name": "e909f8213d" - }, - { - "dataset": "scannetpp", - "scene_name": "e91722b5a3" - }, - { - "dataset": "scannetpp", - "scene_name": "e93cd5230e" - }, - { - "dataset": "scannetpp", - "scene_name": "ed2216380b" - }, - { - "dataset": "scannetpp", - "scene_name": "eea4ad9c04" - }, - { - "dataset": "scannetpp", - "scene_name": "eee7eef8e5" - }, - { - "dataset": "scannetpp", - "scene_name": "ef18cf0708" - }, - { - "dataset": "scannetpp", - "scene_name": "f19ca0a52e" - }, - { - "dataset": "scannetpp", - "scene_name": "f3d64c30f8" - }, - { - "dataset": "scannetpp", - "scene_name": "f3f016ba3f" - }, - { - "dataset": "scannetpp", - "scene_name": "f6659a3107" - }, - { - "dataset": "scannetpp", - "scene_name": "f6a9b64a0d" - }, - { - "dataset": "scannetpp", - "scene_name": "f8d5147d1d" - }, - { - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b" - }, - { - "dataset": "scannetpp", - "scene_name": "fb05e13ad1" - }, - { - "dataset": "scannetpp", - "scene_name": "fe1733741f" - }, - { - "dataset": "scannetpp", - "scene_name": "fe3fc057a1" - } - ] - }, - "qa_pairs": [ - { - "id": "scene0006_00_0", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0006_00_1", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "1", - "question_type": "target presence verification", - "question": "In the entire scene, did the bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0006_00_2", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "2", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bag in this scene?\nA. lamp\nB. backpack\nC. nightstand\nD. shoes", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. shoes" - }, - { - "id": "scene0006_00_3", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "3", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bag is to the side of and above the shoes, about 6 cm away." - }, - { - "id": "scene0006_00_4", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "4", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bag\", \"support_surface\": \"suitcase\", \"references\": [{\"object\": \"shoes\", \"relation\": \"side_above\", \"distance_cm\": 6.3}]}" - }, - { - "id": "scene0006_00_5", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "5", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0006_00_6", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "6", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the purse in this scene?\nA. lamp\nB. bag\nC. backpack\nD. suitcase", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. suitcase" - }, - { - "id": "scene0006_00_7", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the purse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The purse is near the suitcase, about 23 cm away." - }, - { - "id": "scene0006_00_8", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the purse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"purse\", \"support_surface\": \"backpack\", \"references\": [{\"object\": \"suitcase\", \"relation\": \"near\", \"distance_cm\": 23.4}]}" - }, - { - "id": "scene0006_00_9", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "9", - "question_type": "target presence verification", - "question": "In the entire scene, did the toaster oven appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0006_00_10", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "10", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the telephone in this scene?\nA. backpack\nB. lamp\nC. pillow\nD. bag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. lamp" - }, - { - "id": "scene0006_00_11", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "11", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the telephone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The telephone is on the nightstand. It is also to the side of and below the lamp, about 5 cm away. It is further to the side of and below the pillow, about 20 cm away." - }, - { - "id": "scene0006_00_12", - "dataset": "scannet200", - "scene_name": "scene0006_00", - "sample_id": "12", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the telephone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"telephone\", \"support_surface\": \"nightstand\", \"references\": [{\"object\": \"lamp\", \"relation\": \"side_below\", \"distance_cm\": 5.0}, {\"object\": \"pillow\", \"relation\": \"side_below\", \"distance_cm\": 19.7}]}" - }, - { - "id": "scene0024_00_0", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dispenser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0024_00_1", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "1", - "question_type": "target presence verification", - "question": "In the entire scene, did the bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0024_00_2", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "2", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bag in this scene?\nA. book\nB. purse\nC. backpack\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. book" - }, - { - "id": "scene0024_00_3", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "3", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bag is on the table. It is also next to the book, about 15 cm away. It is further to the side of and above the purse, about 59 cm away." - }, - { - "id": "scene0024_00_4", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "4", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bag\", \"support_surface\": \"table\", \"references\": [{\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 14.5}, {\"object\": \"purse\", \"relation\": \"side_above\", \"distance_cm\": 59.1}]}" - }, - { - "id": "scene0024_00_5", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "5", - "question_type": "target presence verification", - "question": "In the entire scene, did the ball appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0024_00_6", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "6", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the ball in this scene?\nA. table\nB. plant\nC. lamp\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "scene0024_00_7", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the ball in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The ball is to the side of and below the table, about 9 cm away. It is also to the side of and below the book, about 31 cm away. It is further to the side of and below the plant, about 36 cm away." - }, - { - "id": "scene0024_00_8", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the ball in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"ball\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 8.7}, {\"object\": \"book\", \"relation\": \"side_below\", \"distance_cm\": 31.0}]}" - }, - { - "id": "scene0024_00_9", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "9", - "question_type": "target presence verification", - "question": "In the entire scene, did the coffee kettle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0024_00_10", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "10", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the purse in this scene?\nA. book\nB. backpack\nC. bag\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. book" - }, - { - "id": "scene0024_00_11", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "11", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the purse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The purse is to the side of and below the book, about 25 cm away. It is also to the side of and below the bag, about 59 cm away." - }, - { - "id": "scene0024_00_12", - "dataset": "scannet200", - "scene_name": "scene0024_00", - "sample_id": "12", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the purse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"purse\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"book\", \"relation\": \"side_below\", \"distance_cm\": 24.6}, {\"object\": \"bag\", \"relation\": \"side_below\", \"distance_cm\": 59.1}]}" - }, - { - "id": "scene0025_00_0", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the book appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0025_00_1", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "1", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"cabinet\", \"references\": []}" - }, - { - "id": "scene0025_00_2", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "2", - "question_type": "target presence verification", - "question": "In the entire scene, did the shoe appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0025_00_3", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "3", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. keyboard\nB. office chair\nC. monitor\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. keyboard" - }, - { - "id": "scene0025_00_4", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "4", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the desk. It is also next to the keyboard, about 3 cm away. It is further to the side of and below the monitor, about 5 cm away." - }, - { - "id": "scene0025_00_5", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "5", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 2.9}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 5.5}]}" - }, - { - "id": "scene0025_00_6", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "6", - "question_type": "target presence verification", - "question": "In the entire scene, did the water pitcher appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0025_00_7", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "7", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper towel roll in this scene?\nA. keyboard\nB. monitor\nC. office chair\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. desk" - }, - { - "id": "scene0025_00_8", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "8", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper towel roll in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper towel roll is to the side of and above the desk, about 13 cm away. It is also next to the keyboard, about 15 cm away. It is further to the side of and above the office chair, about 49 cm away." - }, - { - "id": "scene0025_00_9", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "9", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper towel roll in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper towel roll\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"desk\", \"relation\": \"side_above\", \"distance_cm\": 12.5}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 14.8}]}" - }, - { - "id": "scene0025_00_10", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "10", - "question_type": "target presence verification", - "question": "In the entire scene, did the telephone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0025_00_11", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "11", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the telephone in this scene?\nA. office chair\nB. monitor\nC. box\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. box" - }, - { - "id": "scene0025_00_12", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "12", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the telephone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The telephone is near the monitor, about 22 cm away. It is also to the side of and above the desk, about 28 cm away. It is further to the side of and above the office chair, about 54 cm away." - }, - { - "id": "scene0025_00_13", - "dataset": "scannet200", - "scene_name": "scene0025_00", - "sample_id": "13", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the telephone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"telephone\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"monitor\", \"relation\": \"near\", \"distance_cm\": 21.6}, {\"object\": \"desk\", \"relation\": \"side_above\", \"distance_cm\": 28.0}]}" - }, - { - "id": "scene0034_00_0", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the telephone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0034_00_1", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. mirror\nB. sink\nC. scale\nD. bathroom vanity", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. mirror" - }, - { - "id": "scene0034_00_2", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is to the side of and below the mirror, about 34 cm away." - }, - { - "id": "scene0034_00_3", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"bathroom vanity\", \"references\": [{\"object\": \"mirror\", \"relation\": \"side_below\", \"distance_cm\": 33.9}]}" - }, - { - "id": "scene0034_00_4", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the scale appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0034_00_5", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the scale in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The scale is below the towel, about 17 cm away. It is also to the side of and below the shelf, about 17 cm away." - }, - { - "id": "scene0034_00_6", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the scale in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"scale\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"towel\", \"relation\": \"below\", \"distance_cm\": 17.0}, {\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 17.0}]}" - }, - { - "id": "scene0034_00_7", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the paper bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0034_00_8", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toilet paper in this scene?\nA. toilet\nB. bathroom vanity\nC. trash bin\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. toilet" - }, - { - "id": "scene0034_00_9", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toilet paper in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toilet paper is to the side of and above the toilet, about 43 cm away. It is also to the side of and above the trash bin, about 72 cm away." - }, - { - "id": "scene0034_00_10", - "dataset": "scannet200", - "scene_name": "scene0034_00", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toilet paper in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toilet paper\", \"support_surface\": \"trash bin\", \"references\": [{\"object\": \"toilet\", \"relation\": \"side_above\", \"distance_cm\": 42.7}, {\"object\": \"trash bin\", \"relation\": \"side_above\", \"distance_cm\": 71.9}]}" - }, - { - "id": "scene0040_00_0", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0040_00_1", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the desk. It is also to the side of and above the box, about 11 cm away. It is further to the side of and above the file cabinet, about 12 cm away." - }, - { - "id": "scene0040_00_2", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"box\", \"relation\": \"side_above\", \"distance_cm\": 11.2}, {\"object\": \"file cabinet\", \"relation\": \"side_above\", \"distance_cm\": 12.1}]}" - }, - { - "id": "scene0040_00_3", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "3", - "question_type": "target presence verification", - "question": "In the entire scene, did the paper cutter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0040_00_4", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "4", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the clock in this scene?\nA. bottle\nB. monitor\nC. box\nD. books", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. monitor" - }, - { - "id": "scene0040_00_5", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the clock in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The clock is to the side of and above the monitor, about 53 cm away. It is also to the side of and above the books, about 63 cm away." - }, - { - "id": "scene0040_00_6", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the clock in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"clock\", \"support_surface\": \"books\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_above\", \"distance_cm\": 52.5}, {\"object\": \"books\", \"relation\": \"side_above\", \"distance_cm\": 63.5}]}" - }, - { - "id": "scene0040_00_7", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dish appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0040_00_8", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. file cabinet\nB. teddy bear\nC. keyboard\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. keyboard" - }, - { - "id": "scene0040_00_9", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the desk. It is also to the side of and above the keyboard, about 12 cm away. It is further near the cup, about 17 cm away." - }, - { - "id": "scene0040_00_10", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"side_above\", \"distance_cm\": 11.9}, {\"object\": \"cup\", \"relation\": \"near\", \"distance_cm\": 16.9}]}" - }, - { - "id": "scene0040_00_11", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0040_00_12", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. file cabinet\nB. box\nC. bottle\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. box" - }, - { - "id": "scene0040_00_13", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also to the side of and above the box, about 22 cm away. It is further near the bottle, about 29 cm away." - }, - { - "id": "scene0040_00_14", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"box\", \"relation\": \"side_above\", \"distance_cm\": 22.4}, {\"object\": \"bottle\", \"relation\": \"near\", \"distance_cm\": 28.5}]}" - }, - { - "id": "scene0040_00_15", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the cd case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0040_00_16", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper bag in this scene?\nA. box\nB. light switch\nC. books\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. light switch" - }, - { - "id": "scene0040_00_17", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper bag is to the side of and below the light switch, about 69 cm away." - }, - { - "id": "scene0040_00_18", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper bag\", \"support_surface\": \"light switch\", \"references\": [{\"object\": \"light switch\", \"relation\": \"side_below\", \"distance_cm\": 68.6}]}" - }, - { - "id": "scene0040_00_19", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "19", - "question_type": "target presence verification", - "question": "In the entire scene, did the ball appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0040_00_20", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "20", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the stuffed animal in this scene?\nA. file cabinet\nB. monitor\nC. box\nD. books", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. books" - }, - { - "id": "scene0040_00_21", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "21", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the stuffed animal in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The stuffed animal is on the desk. It is also next to the books, about 9 cm away. It is further to the side of and above the file cabinet, about 37 cm away." - }, - { - "id": "scene0040_00_22", - "dataset": "scannet200", - "scene_name": "scene0040_00", - "sample_id": "22", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the stuffed animal in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"stuffed animal\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"books\", \"relation\": \"next_to\", \"distance_cm\": 8.8}, {\"object\": \"file cabinet\", \"relation\": \"side_above\", \"distance_cm\": 36.6}]}" - }, - { - "id": "scene0042_00_0", - "dataset": "scannet200", - "scene_name": "scene0042_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the paper cutter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0042_00_1", - "dataset": "scannet200", - "scene_name": "scene0042_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dispenser in this scene?\nA. chair\nB. mirror\nC. sink\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. sink" - }, - { - "id": "scene0042_00_2", - "dataset": "scannet200", - "scene_name": "scene0042_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dispenser is to the side of and above the sink, about 29 cm away. It is also near the mirror, about 48 cm away." - }, - { - "id": "scene0042_00_3", - "dataset": "scannet200", - "scene_name": "scene0042_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dispenser\", \"support_surface\": \"sink\", \"references\": [{\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 28.6}, {\"object\": \"mirror\", \"relation\": \"near\", \"distance_cm\": 47.7}]}" - }, - { - "id": "scene0047_00_0", - "dataset": "scannet200", - "scene_name": "scene0047_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the hat appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0047_00_1", - "dataset": "scannet200", - "scene_name": "scene0047_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dispenser in this scene?\nA. microwave\nB. kitchen cabinets\nC. sink\nD. paper towel dispenser", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. paper towel dispenser" - }, - { - "id": "scene0047_00_2", - "dataset": "scannet200", - "scene_name": "scene0047_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dispenser is next to the paper towel dispenser, about 6 cm away. It is also to the side of and above the sink, about 19 cm away. It is further to the side of and below the kitchen cabinets, about 26 cm away." - }, - { - "id": "scene0047_00_3", - "dataset": "scannet200", - "scene_name": "scene0047_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dispenser\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"paper towel dispenser\", \"relation\": \"next_to\", \"distance_cm\": 5.5}, {\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 19.4}]}" - }, - { - "id": "scene0048_00_0", - "dataset": "scannet200", - "scene_name": "scene0048_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the clock appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0048_00_1", - "dataset": "scannet200", - "scene_name": "scene0048_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. chair\nB. monitor\nC. table\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "scene0048_00_2", - "dataset": "scannet200", - "scene_name": "scene0048_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the table. It is also to the side of and above the chair, about 54 cm away." - }, - { - "id": "scene0048_00_3", - "dataset": "scannet200", - "scene_name": "scene0048_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 53.7}]}" - }, - { - "id": "scene0070_00_0", - "dataset": "scannet200", - "scene_name": "scene0070_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the clock appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0070_00_1", - "dataset": "scannet200", - "scene_name": "scene0070_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the clock in this scene?\nA. chair\nB. picture\nC. file cabinet\nD. ledge", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. picture" - }, - { - "id": "scene0070_00_2", - "dataset": "scannet200", - "scene_name": "scene0070_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the clock in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The clock is to the side of and above the picture, about 13 cm away. It is also to the side of and above the ledge, about 67 cm away." - }, - { - "id": "scene0070_00_3", - "dataset": "scannet200", - "scene_name": "scene0070_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the clock in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"clock\", \"support_surface\": \"picture\", \"references\": [{\"object\": \"picture\", \"relation\": \"side_above\", \"distance_cm\": 13.1}, {\"object\": \"ledge\", \"relation\": \"side_above\", \"distance_cm\": 67.1}]}" - }, - { - "id": "scene0070_00_4", - "dataset": "scannet200", - "scene_name": "scene0070_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0070_00_5", - "dataset": "scannet200", - "scene_name": "scene0070_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. monitor\nB. chair\nC. file cabinet\nD. ledge", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. file cabinet" - }, - { - "id": "scene0070_00_6", - "dataset": "scannet200", - "scene_name": "scene0070_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also to the side of and above the file cabinet, about 18 cm away. It is further to the side of and below the monitor, about 26 cm away." - }, - { - "id": "scene0070_00_7", - "dataset": "scannet200", - "scene_name": "scene0070_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"file cabinet\", \"relation\": \"side_above\", \"distance_cm\": 18.0}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 25.8}]}" - }, - { - "id": "scene0083_00_0", - "dataset": "scannet200", - "scene_name": "scene0083_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0083_00_1", - "dataset": "scannet200", - "scene_name": "scene0083_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dish in this scene?\nA. monitor\nB. toilet paper\nC. toilet\nD. bathtub", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bathtub" - }, - { - "id": "scene0083_00_2", - "dataset": "scannet200", - "scene_name": "scene0083_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dish in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dish is to the side of and above the bathtub, about 8 cm away." - }, - { - "id": "scene0083_00_3", - "dataset": "scannet200", - "scene_name": "scene0083_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dish in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dish\", \"support_surface\": \"bathtub\", \"references\": [{\"object\": \"bathtub\", \"relation\": \"side_above\", \"distance_cm\": 8.3}]}" - }, - { - "id": "scene0083_00_4", - "dataset": "scannet200", - "scene_name": "scene0083_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the toilet paper appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0083_00_5", - "dataset": "scannet200", - "scene_name": "scene0083_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toilet paper in this scene?\nA. monitor\nB. toilet\nC. bathtub\nD. soap dish", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. toilet" - }, - { - "id": "scene0083_00_6", - "dataset": "scannet200", - "scene_name": "scene0083_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toilet paper in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toilet paper is to the side of and above the toilet, about 22 cm away." - }, - { - "id": "scene0083_00_7", - "dataset": "scannet200", - "scene_name": "scene0083_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toilet paper in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toilet paper\", \"support_surface\": \"toilet\", \"references\": [{\"object\": \"toilet\", \"relation\": \"side_above\", \"distance_cm\": 21.9}]}" - }, - { - "id": "scene0084_00_0", - "dataset": "scannet200", - "scene_name": "scene0084_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0084_00_1", - "dataset": "scannet200", - "scene_name": "scene0084_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dispenser in this scene?\nA. mirror\nB. bathroom counter\nC. sink\nD. sign", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. bathroom counter" - }, - { - "id": "scene0084_00_2", - "dataset": "scannet200", - "scene_name": "scene0084_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dispenser is to the side of and above the bathroom counter, about 3 cm away. It is also to the side of and above the sink, about 19 cm away. It is further near the mirror, about 42 cm away." - }, - { - "id": "scene0084_00_3", - "dataset": "scannet200", - "scene_name": "scene0084_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dispenser\", \"support_surface\": \"bathroom counter\", \"references\": [{\"object\": \"bathroom counter\", \"relation\": \"side_above\", \"distance_cm\": 3.2}, {\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 19.2}]}" - }, - { - "id": "scene0084_00_4", - "dataset": "scannet200", - "scene_name": "scene0084_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the toilet paper appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0084_00_5", - "dataset": "scannet200", - "scene_name": "scene0084_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toilet paper in this scene?\nA. bathroom counter\nB. jacket\nC. toilet\nD. mirror", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. toilet" - }, - { - "id": "scene0084_00_6", - "dataset": "scannet200", - "scene_name": "scene0084_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toilet paper in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toilet paper is to the side of and above the toilet, about 28 cm away." - }, - { - "id": "scene0084_00_7", - "dataset": "scannet200", - "scene_name": "scene0084_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toilet paper in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toilet paper\", \"support_surface\": \"toilet paper holder\", \"references\": [{\"object\": \"toilet\", \"relation\": \"side_above\", \"distance_cm\": 27.9}]}" - }, - { - "id": "scene0086_00_0", - "dataset": "scannet200", - "scene_name": "scene0086_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0086_00_1", - "dataset": "scannet200", - "scene_name": "scene0086_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. trash can\nB. paper towel dispenser\nC. plant\nD. mirror", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. trash can" - }, - { - "id": "scene0086_00_2", - "dataset": "scannet200", - "scene_name": "scene0086_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the trash can, about 24 cm away. It is also to the side of and above the paper towel dispenser, about 68 cm away." - }, - { - "id": "scene0086_00_3", - "dataset": "scannet200", - "scene_name": "scene0086_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"paper towel dispenser\", \"references\": [{\"object\": \"trash can\", \"relation\": \"side_above\", \"distance_cm\": 24.1}, {\"object\": \"paper towel dispenser\", \"relation\": \"side_above\", \"distance_cm\": 68.0}]}" - }, - { - "id": "scene0086_00_4", - "dataset": "scannet200", - "scene_name": "scene0086_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the toilet paper appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0086_00_5", - "dataset": "scannet200", - "scene_name": "scene0086_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dispenser in this scene?\nA. bottle\nB. plant\nC. sink\nD. paper towel dispenser", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. sink" - }, - { - "id": "scene0086_00_6", - "dataset": "scannet200", - "scene_name": "scene0086_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dispenser is to the side of and above the sink, about 7 cm away. It is also to the side of and above the paper towel dispenser, about 44 cm away. It is further to the side of and below the plant, about 69 cm away." - }, - { - "id": "scene0086_00_7", - "dataset": "scannet200", - "scene_name": "scene0086_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dispenser\", \"support_surface\": \"paper towel dispenser\", \"references\": [{\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 7.0}, {\"object\": \"paper towel dispenser\", \"relation\": \"side_above\", \"distance_cm\": 43.5}]}" - }, - { - "id": "scene0089_00_0", - "dataset": "scannet200", - "scene_name": "scene0089_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0089_00_1", - "dataset": "scannet200", - "scene_name": "scene0089_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. desk\nB. monitor\nC. radiator\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "scene0089_00_2", - "dataset": "scannet200", - "scene_name": "scene0089_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is to the side of and above the chair, about 7 cm away." - }, - { - "id": "scene0089_00_3", - "dataset": "scannet200", - "scene_name": "scene0089_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 7.5}]}" - }, - { - "id": "scene0095_00_0", - "dataset": "scannet200", - "scene_name": "scene0095_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0095_00_1", - "dataset": "scannet200", - "scene_name": "scene0095_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. chair\nB. desk\nC. mouse\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. monitor" - }, - { - "id": "scene0095_00_2", - "dataset": "scannet200", - "scene_name": "scene0095_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is to the side of and below the monitor, about 9 cm away. It is also next to the mouse, about 12 cm away. It is further to the side of and above the chair, about 21 cm away." - }, - { - "id": "scene0095_00_3", - "dataset": "scannet200", - "scene_name": "scene0095_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 9.0}, {\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 12.2}]}" - }, - { - "id": "scene0095_00_4", - "dataset": "scannet200", - "scene_name": "scene0095_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0095_00_5", - "dataset": "scannet200", - "scene_name": "scene0095_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. monitor\nB. keyboard\nC. desk\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "scene0095_00_6", - "dataset": "scannet200", - "scene_name": "scene0095_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is to the side of and above the chair, about 11 cm away. It is also next to the keyboard, about 12 cm away. It is further to the side of and below the monitor, about 26 cm away." - }, - { - "id": "scene0095_00_7", - "dataset": "scannet200", - "scene_name": "scene0095_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 11.1}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 12.2}]}" - }, - { - "id": "scene0097_00_0", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the plate appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0097_00_1", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. coffee maker\nB. cabinet\nC. coffee kettle\nD. counter", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cabinet" - }, - { - "id": "scene0097_00_2", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the cabinet, about 34 cm away." - }, - { - "id": "scene0097_00_3", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 33.6}]}" - }, - { - "id": "scene0097_00_4", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the coffee kettle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0097_00_5", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the coffee kettle in this scene?\nA. toaster\nB. sink\nC. kitchen cabinets\nD. coffee maker", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. toaster" - }, - { - "id": "scene0097_00_6", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the coffee kettle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The coffee kettle is on the counter. It is also under the toaster, about 3 cm away. It is further to the side of and above the kitchen cabinets, about 13 cm away." - }, - { - "id": "scene0097_00_7", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the coffee kettle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"coffee kettle\", \"support_surface\": \"counter\", \"references\": [{\"object\": \"toaster\", \"relation\": \"under\", \"distance_cm\": 3.4}, {\"object\": \"kitchen cabinets\", \"relation\": \"side_above\", \"distance_cm\": 13.1}]}" - }, - { - "id": "scene0097_00_8", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the scale appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0097_00_9", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toaster in this scene?\nA. sink\nB. coffee maker\nC. coffee kettle\nD. soap", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. coffee kettle" - }, - { - "id": "scene0097_00_10", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toaster in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toaster is on the counter. It is also next to the coffee kettle, about 3 cm away. It is further to the side of and below the soap, about 6 cm away." - }, - { - "id": "scene0097_00_11", - "dataset": "scannet200", - "scene_name": "scene0097_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toaster in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toaster\", \"support_surface\": \"counter\", \"references\": [{\"object\": \"coffee kettle\", \"relation\": \"next_to\", \"distance_cm\": 3.4}, {\"object\": \"soap\", \"relation\": \"side_below\", \"distance_cm\": 5.9}]}" - }, - { - "id": "scene0100_00_0", - "dataset": "scannet200", - "scene_name": "scene0100_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0100_00_1", - "dataset": "scannet200", - "scene_name": "scene0100_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toilet paper in this scene?\nA. shower control valve\nB. bathtub\nC. toilet\nD. towel", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. toilet" - }, - { - "id": "scene0100_00_2", - "dataset": "scannet200", - "scene_name": "scene0100_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toilet paper in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toilet paper is to the side of and above the toilet, about 15 cm away. It is also to the side of and above the bathtub, about 27 cm away. It is further to the side of and below the shower control valve, about 60 cm away." - }, - { - "id": "scene0100_00_3", - "dataset": "scannet200", - "scene_name": "scene0100_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toilet paper in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toilet paper\", \"support_surface\": \"toilet\", \"references\": [{\"object\": \"toilet\", \"relation\": \"side_above\", \"distance_cm\": 15.5}, {\"object\": \"bathtub\", \"relation\": \"side_above\", \"distance_cm\": 26.9}]}" - }, - { - "id": "scene0101_00_0", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0101_00_1", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. backpack\nB. laptop\nC. bed\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. bed" - }, - { - "id": "scene0101_00_2", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the desk. It is also to the side of and above the bed, about 48 cm away. It is further to the side of and above the chair, about 62 cm away." - }, - { - "id": "scene0101_00_3", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"bed\", \"relation\": \"side_above\", \"distance_cm\": 47.9}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 61.7}]}" - }, - { - "id": "scene0101_00_4", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0101_00_5", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the dumbbell in this scene?\nA. bed\nB. towel\nC. backpack\nD. bathroom vanity", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. towel" - }, - { - "id": "scene0101_00_6", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the dumbbell in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The dumbbell is below the towel, about 41 cm away." - }, - { - "id": "scene0101_00_7", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the dumbbell in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"dumbbell\", \"support_surface\": \"stove\", \"references\": [{\"object\": \"towel\", \"relation\": \"below\", \"distance_cm\": 41.2}]}" - }, - { - "id": "scene0101_00_8", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the cd case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0101_00_9", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "9", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dispenser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0101_00_10", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "10", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the laptop in this scene?\nA. cup\nB. trash can\nC. shelf\nD. backpack", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. shelf" - }, - { - "id": "scene0101_00_11", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "11", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the laptop in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The laptop is on the desk. It is also to the side of and below the shelf, about 24 cm away. It is further to the side of and above the trash can, about 65 cm away." - }, - { - "id": "scene0101_00_12", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "12", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the laptop in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"laptop\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 24.2}, {\"object\": \"trash can\", \"relation\": \"side_above\", \"distance_cm\": 65.4}]}" - }, - { - "id": "scene0101_00_13", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "13", - "question_type": "target presence verification", - "question": "In the entire scene, did the stuffed animal appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0101_00_14", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "14", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the plunger in this scene?\nA. bathroom vanity\nB. toilet paper\nC. backpack\nD. toilet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. toilet" - }, - { - "id": "scene0101_00_15", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "15", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the plunger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The plunger is to the side of and below the toilet, about 20 cm away. It is also to the side of and below the toilet paper, about 74 cm away." - }, - { - "id": "scene0101_00_16", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "16", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the plunger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"plunger\", \"support_surface\": \"toilet paper\", \"references\": [{\"object\": \"toilet\", \"relation\": \"side_below\", \"distance_cm\": 20.4}, {\"object\": \"toilet paper\", \"relation\": \"side_below\", \"distance_cm\": 74.1}]}" - }, - { - "id": "scene0101_00_17", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "17", - "question_type": "target presence verification", - "question": "In the entire scene, did the toilet paper appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0101_00_18", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "18", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toilet paper in this scene?\nA. backpack\nB. toilet\nC. sink\nD. plunger", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. toilet" - }, - { - "id": "scene0101_00_19", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "19", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toilet paper in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toilet paper is to the side of and above the toilet, about 16 cm away. It is also to the side of and below the sink, about 31 cm away. It is further to the side of and above the plunger, about 74 cm away." - }, - { - "id": "scene0101_00_20", - "dataset": "scannet200", - "scene_name": "scene0101_00", - "sample_id": "20", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toilet paper in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toilet paper\", \"support_surface\": \"bathroom vanity\", \"references\": [{\"object\": \"toilet\", \"relation\": \"side_above\", \"distance_cm\": 15.6}, {\"object\": \"sink\", \"relation\": \"side_below\", \"distance_cm\": 30.6}]}" - }, - { - "id": "scene0109_00_0", - "dataset": "scannet200", - "scene_name": "scene0109_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the water pitcher appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0117_00_0", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0117_00_1", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. cup\nB. tupperware\nC. cups\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. tupperware" - }, - { - "id": "scene0117_00_2", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the tupperware, about 42 cm away." - }, - { - "id": "scene0117_00_3", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"tupperware\", \"relation\": \"side_below\", \"distance_cm\": 42.3}]}" - }, - { - "id": "scene0117_00_4", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the broom appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0117_00_5", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper towel roll in this scene?\nA. kitchen cabinets\nB. cup\nC. toaster\nD. cups", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. kitchen cabinets" - }, - { - "id": "scene0117_00_6", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper towel roll in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper towel roll is to the side of and above the kitchen cabinets, about 18 cm away. It is also to the side of and below the cup, about 28 cm away. It is further to the side of and below the cups, about 33 cm away." - }, - { - "id": "scene0117_00_7", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper towel roll in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper towel roll\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"kitchen cabinets\", \"relation\": \"side_above\", \"distance_cm\": 18.2}, {\"object\": \"cup\", \"relation\": \"side_below\", \"distance_cm\": 28.3}]}" - }, - { - "id": "scene0117_00_8", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the guitar appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0117_00_9", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dispenser in this scene?\nA. chair\nB. backpack\nC. sink\nD. paper towel dispenser", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. paper towel dispenser" - }, - { - "id": "scene0117_00_10", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dispenser is to the side of and below the paper towel dispenser, about 63 cm away. It is also to the side of and above the sink, about 65 cm away." - }, - { - "id": "scene0117_00_11", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dispenser\", \"support_surface\": \"paper towel dispenser\", \"references\": [{\"object\": \"paper towel dispenser\", \"relation\": \"side_below\", \"distance_cm\": 62.6}, {\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 64.9}]}" - }, - { - "id": "scene0117_00_12", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the toaster appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0117_00_13", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toaster in this scene?\nA. refrigerator\nB. cup\nC. cups\nD. paper towel roll", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. cups" - }, - { - "id": "scene0117_00_14", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toaster in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toaster is on the shelf. It is also to the side of and below the cups, about 29 cm away. It is further near the refrigerator, about 36 cm away." - }, - { - "id": "scene0117_00_15", - "dataset": "scannet200", - "scene_name": "scene0117_00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toaster in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toaster\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"cups\", \"relation\": \"side_below\", \"distance_cm\": 28.9}, {\"object\": \"refrigerator\", \"relation\": \"near\", \"distance_cm\": 36.4}]}" - }, - { - "id": "scene0121_00_0", - "dataset": "scannet200", - "scene_name": "scene0121_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the water pitcher appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0121_00_1", - "dataset": "scannet200", - "scene_name": "scene0121_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the purse in this scene?\nA. clothes\nB. mirror\nC. chair\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. mirror" - }, - { - "id": "scene0121_00_2", - "dataset": "scannet200", - "scene_name": "scene0121_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the purse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The purse is to the side of and above the mirror, about 15 cm away." - }, - { - "id": "scene0121_00_3", - "dataset": "scannet200", - "scene_name": "scene0121_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the purse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"purse\", \"support_surface\": \"clothes\", \"references\": [{\"object\": \"mirror\", \"relation\": \"side_above\", \"distance_cm\": 14.6}]}" - }, - { - "id": "scene0136_00_0", - "dataset": "scannet200", - "scene_name": "scene0136_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the candle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0136_00_1", - "dataset": "scannet200", - "scene_name": "scene0136_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper cutter in this scene?\nA. cabinets\nB. paper\nC. cabinet\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. paper" - }, - { - "id": "scene0136_00_2", - "dataset": "scannet200", - "scene_name": "scene0136_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper cutter in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper cutter is under the paper, about 5 cm away. It is also to the side of and below the cabinet, about 48 cm away. It is further to the side of and above the box, about 56 cm away." - }, - { - "id": "scene0136_00_3", - "dataset": "scannet200", - "scene_name": "scene0136_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper cutter in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper cutter\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"paper\", \"relation\": \"under\", \"distance_cm\": 4.6}, {\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 47.8}]}" - }, - { - "id": "scene0140_00_0", - "dataset": "scannet200", - "scene_name": "scene0140_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0156_00_0", - "dataset": "scannet200", - "scene_name": "scene0156_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the dustpan appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0156_00_1", - "dataset": "scannet200", - "scene_name": "scene0156_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. desk\nB. keyboard\nC. monitor\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "scene0156_00_2", - "dataset": "scannet200", - "scene_name": "scene0156_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the chair, about 51 cm away." - }, - { - "id": "scene0156_00_3", - "dataset": "scannet200", - "scene_name": "scene0156_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 50.7}]}" - }, - { - "id": "scene0164_00_0", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the scale appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0164_00_1", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. beer bottles\nB. bag\nC. box\nD. card", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. bag" - }, - { - "id": "scene0164_00_2", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is near the bag, about 57 cm away." - }, - { - "id": "scene0164_00_3", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"bag\", \"references\": [{\"object\": \"bag\", \"relation\": \"near\", \"distance_cm\": 57.3}]}" - }, - { - "id": "scene0164_00_4", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dish appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0164_00_5", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bag in this scene?\nA. box\nB. kitchen cabinet\nC. backpack\nD. beer bottles", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. kitchen cabinet" - }, - { - "id": "scene0164_00_6", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bag is to the side of and below the kitchen cabinet, about 41 cm away. It is also near the backpack, about 57 cm away." - }, - { - "id": "scene0164_00_7", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bag\", \"support_surface\": \"kitchen cabinet\", \"references\": [{\"object\": \"kitchen cabinet\", \"relation\": \"side_below\", \"distance_cm\": 40.9}, {\"object\": \"backpack\", \"relation\": \"near\", \"distance_cm\": 57.3}]}" - }, - { - "id": "scene0164_00_8", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the coffee kettle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0164_00_9", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the coffee kettle in this scene?\nA. dishwasher\nB. picture\nC. dish rack\nD. faucet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. faucet" - }, - { - "id": "scene0164_00_10", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the coffee kettle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The coffee kettle is on the counter. It is also to the side of and above the faucet, about 4 cm away. It is further to the side of and above the dishwasher, about 36 cm away." - }, - { - "id": "scene0164_00_11", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the coffee kettle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"coffee kettle\", \"support_surface\": \"counter\", \"references\": [{\"object\": \"faucet\", \"relation\": \"side_above\", \"distance_cm\": 3.6}, {\"object\": \"dishwasher\", \"relation\": \"side_above\", \"distance_cm\": 36.2}]}" - }, - { - "id": "scene0164_00_12", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the hair dryer appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0164_00_13", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper towel roll in this scene?\nA. soda stream\nB. beer bottles\nC. kitchen cabinets\nD. kitchen cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. kitchen cabinet" - }, - { - "id": "scene0164_00_14", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper towel roll in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper towel roll is to the side of and below the kitchen cabinet, about 12 cm away. It is also to the side of and above the kitchen cabinets, about 18 cm away. It is further near the soda stream, about 20 cm away." - }, - { - "id": "scene0164_00_15", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper towel roll in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper towel roll\", \"support_surface\": \"kitchen cabinet\", \"references\": [{\"object\": \"kitchen cabinet\", \"relation\": \"side_below\", \"distance_cm\": 12.5}, {\"object\": \"kitchen cabinets\", \"relation\": \"side_above\", \"distance_cm\": 18.0}]}" - }, - { - "id": "scene0164_00_16", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the toaster appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0164_00_17", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "17", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toaster in this scene?\nA. microwave\nB. open kitchen cabinet\nC. kitchen cabinet\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. microwave" - }, - { - "id": "scene0164_00_18", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "18", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toaster in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toaster is next to the microwave, about 4 cm away. It is also to the side of and below the open kitchen cabinet, about 16 cm away. It is further to the side of and below the kitchen cabinet, about 23 cm away." - }, - { - "id": "scene0164_00_19", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "19", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toaster in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toaster\", \"support_surface\": \"open kitchen cabinet\", \"references\": [{\"object\": \"microwave\", \"relation\": \"next_to\", \"distance_cm\": 4.5}, {\"object\": \"open kitchen cabinet\", \"relation\": \"side_below\", \"distance_cm\": 16.4}]}" - }, - { - "id": "scene0164_00_20", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "20", - "question_type": "target presence verification", - "question": "In the entire scene, did the telephone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0164_00_21", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "21", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water bottle is to the side of and above the refrigerator, about 6 cm away. It is also next to the carton, about 6 cm away. It is further to the side of and above the kitchen cabinets, about 12 cm away." - }, - { - "id": "scene0164_00_22", - "dataset": "scannet200", - "scene_name": "scene0164_00", - "sample_id": "22", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water bottle\", \"support_surface\": \"kitchen cabinet\", \"references\": [{\"object\": \"refrigerator\", \"relation\": \"side_above\", \"distance_cm\": 5.8}, {\"object\": \"carton\", \"relation\": \"next_to\", \"distance_cm\": 5.8}]}" - }, - { - "id": "scene0166_00_0", - "dataset": "scannet200", - "scene_name": "scene0166_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the ball appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0166_00_1", - "dataset": "scannet200", - "scene_name": "scene0166_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the ball in this scene?\nA. bag\nB. book\nC. end table\nD. stool", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. stool" - }, - { - "id": "scene0166_00_2", - "dataset": "scannet200", - "scene_name": "scene0166_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the ball in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The ball is to the side of and below the stool, about 10 cm away." - }, - { - "id": "scene0166_00_3", - "dataset": "scannet200", - "scene_name": "scene0166_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the ball in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"ball\", \"support_surface\": \"stool\", \"references\": [{\"object\": \"stool\", \"relation\": \"side_below\", \"distance_cm\": 9.6}]}" - }, - { - "id": "scene0166_00_4", - "dataset": "scannet200", - "scene_name": "scene0166_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0166_00_5", - "dataset": "scannet200", - "scene_name": "scene0166_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the book in this scene?\nA. thermostat\nB. end table\nC. pillow\nD. bag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. end table" - }, - { - "id": "scene0166_00_6", - "dataset": "scannet200", - "scene_name": "scene0166_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the book in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The book is near the bag, about 31 cm away. It is also to the side of and below the pillow, about 67 cm away. It is further to the side of and below the thermostat, about 72 cm away." - }, - { - "id": "scene0166_00_7", - "dataset": "scannet200", - "scene_name": "scene0166_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"end table\", \"references\": [{\"object\": \"bag\", \"relation\": \"near\", \"distance_cm\": 31.1}, {\"object\": \"pillow\", \"relation\": \"side_below\", \"distance_cm\": 67.0}]}" - }, - { - "id": "scene0188_00_0", - "dataset": "scannet200", - "scene_name": "scene0188_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the toaster oven appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0188_00_1", - "dataset": "scannet200", - "scene_name": "scene0188_00", - "sample_id": "1", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"cabinet\", \"references\": []}" - }, - { - "id": "scene0191_00_0", - "dataset": "scannet200", - "scene_name": "scene0191_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the book appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0191_00_1", - "dataset": "scannet200", - "scene_name": "scene0191_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. chair\nB. desk\nC. monitor\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "scene0191_00_2", - "dataset": "scannet200", - "scene_name": "scene0191_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the table. It is also to the side of and above the chair, about 35 cm away." - }, - { - "id": "scene0191_00_3", - "dataset": "scannet200", - "scene_name": "scene0191_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 35.0}]}" - }, - { - "id": "scene0196_00_0", - "dataset": "scannet200", - "scene_name": "scene0196_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the plunger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0196_00_1", - "dataset": "scannet200", - "scene_name": "scene0196_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. table\nB. desk\nC. monitor\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "scene0196_00_2", - "dataset": "scannet200", - "scene_name": "scene0196_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the table. It is also to the side of and above the chair, about 33 cm away." - }, - { - "id": "scene0196_00_3", - "dataset": "scannet200", - "scene_name": "scene0196_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 33.2}]}" - }, - { - "id": "scene0207_00_0", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the paper bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0207_00_1", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper towel roll in this scene?\nA. desk\nB. nightstand\nC. plate\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. plate" - }, - { - "id": "scene0207_00_2", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper towel roll in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper towel roll is on the desk. It is also near the plate, about 15 cm away. It is further to the side of and below the monitor, about 50 cm away." - }, - { - "id": "scene0207_00_3", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper towel roll in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper towel roll\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"plate\", \"relation\": \"near\", \"distance_cm\": 15.1}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 50.1}]}" - }, - { - "id": "scene0207_00_4", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0207_00_5", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the plate in this scene?\nA. monitor\nB. nightstand\nC. desk\nD. paper towel roll", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. paper towel roll" - }, - { - "id": "scene0207_00_6", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the plate in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The plate is on the desk. It is also near the paper towel roll, about 15 cm away. It is further to the side of and below the monitor, about 25 cm away." - }, - { - "id": "scene0207_00_7", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the plate in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"plate\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"paper towel roll\", \"relation\": \"near\", \"distance_cm\": 15.1}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 25.1}]}" - }, - { - "id": "scene0207_00_8", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue box appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0207_00_9", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. shoes\nB. nightstand\nC. monitor\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. monitor" - }, - { - "id": "scene0207_00_10", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is on the desk. It is also to the side of and below the monitor, about 26 cm away. It is further to the side of and above the nightstand, about 49 cm away." - }, - { - "id": "scene0207_00_11", - "dataset": "scannet200", - "scene_name": "scene0207_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 26.1}, {\"object\": \"nightstand\", \"relation\": \"side_above\", \"distance_cm\": 49.4}]}" - }, - { - "id": "scene0211_00_0", - "dataset": "scannet200", - "scene_name": "scene0211_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the tray appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0212_00_0", - "dataset": "scannet200", - "scene_name": "scene0212_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dispenser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0212_00_1", - "dataset": "scannet200", - "scene_name": "scene0212_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. chair\nB. monitor\nC. banister\nD. staircase", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. staircase" - }, - { - "id": "scene0212_00_2", - "dataset": "scannet200", - "scene_name": "scene0212_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the staircase, about 18 cm away." - }, - { - "id": "scene0212_00_3", - "dataset": "scannet200", - "scene_name": "scene0212_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"banister\", \"references\": [{\"object\": \"staircase\", \"relation\": \"side_below\", \"distance_cm\": 17.9}]}" - }, - { - "id": "scene0213_00_0", - "dataset": "scannet200", - "scene_name": "scene0213_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the toaster oven appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0213_00_1", - "dataset": "scannet200", - "scene_name": "scene0213_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. coffee table\nB. armchair\nC. bag\nD. bench", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bench" - }, - { - "id": "scene0213_00_2", - "dataset": "scannet200", - "scene_name": "scene0213_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the bench, about 26 cm away." - }, - { - "id": "scene0213_00_3", - "dataset": "scannet200", - "scene_name": "scene0213_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"coffee table\", \"references\": [{\"object\": \"bench\", \"relation\": \"side_below\", \"distance_cm\": 26.4}]}" - }, - { - "id": "scene0213_00_4", - "dataset": "scannet200", - "scene_name": "scene0213_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0213_00_5", - "dataset": "scannet200", - "scene_name": "scene0213_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bag in this scene?\nA. shelf\nB. armchair\nC. windowsill\nD. lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. lamp" - }, - { - "id": "scene0213_00_6", - "dataset": "scannet200", - "scene_name": "scene0213_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bag is under the lamp. It is also to the side of and below the armchair, about 14 cm away. It is further to the side of and below the windowsill, about 34 cm away." - }, - { - "id": "scene0213_00_7", - "dataset": "scannet200", - "scene_name": "scene0213_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bag\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"lamp\", \"relation\": \"under\", \"distance_cm\": 2.1}, {\"object\": \"armchair\", \"relation\": \"side_below\", \"distance_cm\": 14.0}]}" - }, - { - "id": "scene0220_00_0", - "dataset": "scannet200", - "scene_name": "scene0220_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0222_00_0", - "dataset": "scannet200", - "scene_name": "scene0222_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the dustpan appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0222_00_1", - "dataset": "scannet200", - "scene_name": "scene0222_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. desk\nB. shoes\nC. chair\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. shoes" - }, - { - "id": "scene0222_00_2", - "dataset": "scannet200", - "scene_name": "scene0222_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the shoes, about 25 cm away." - }, - { - "id": "scene0222_00_3", - "dataset": "scannet200", - "scene_name": "scene0222_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"shoes\", \"references\": [{\"object\": \"shoes\", \"relation\": \"side_above\", \"distance_cm\": 25.0}]}" - }, - { - "id": "scene0225_00_0", - "dataset": "scannet200", - "scene_name": "scene0225_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bowl appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0225_00_1", - "dataset": "scannet200", - "scene_name": "scene0225_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. cabinet\nB. monitor\nC. chair\nD. trash can", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. trash can" - }, - { - "id": "scene0225_00_2", - "dataset": "scannet200", - "scene_name": "scene0225_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the cabinet. It is also to the side of and above the trash can, about 17 cm away." - }, - { - "id": "scene0225_00_3", - "dataset": "scannet200", - "scene_name": "scene0225_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"trash can\", \"relation\": \"side_above\", \"distance_cm\": 17.4}]}" - }, - { - "id": "scene0226_00_0", - "dataset": "scannet200", - "scene_name": "scene0226_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0226_00_1", - "dataset": "scannet200", - "scene_name": "scene0226_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the book in this scene?\nA. monitor\nB. sofa chair\nC. lamp\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. sofa chair" - }, - { - "id": "scene0226_00_2", - "dataset": "scannet200", - "scene_name": "scene0226_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the book in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The book is on the table. It is also to the side of and above the sofa chair, about 17 cm away. It is further to the side of and below the lamp, about 63 cm away." - }, - { - "id": "scene0226_00_3", - "dataset": "scannet200", - "scene_name": "scene0226_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"table\", \"references\": [{\"object\": \"sofa chair\", \"relation\": \"side_above\", \"distance_cm\": 16.6}, {\"object\": \"lamp\", \"relation\": \"side_below\", \"distance_cm\": 63.0}]}" - }, - { - "id": "scene0233_00_0", - "dataset": "scannet200", - "scene_name": "scene0233_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the stuffed animal appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0233_00_1", - "dataset": "scannet200", - "scene_name": "scene0233_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. trash can\nB. bag\nC. cabinet\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. trash can" - }, - { - "id": "scene0233_00_2", - "dataset": "scannet200", - "scene_name": "scene0233_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the trash can, about 17 cm away. It is also to the side of and below the cabinet, about 42 cm away." - }, - { - "id": "scene0233_00_3", - "dataset": "scannet200", - "scene_name": "scene0233_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"trash can\", \"relation\": \"side_above\", \"distance_cm\": 16.9}, {\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 41.8}]}" - }, - { - "id": "scene0252_00_0", - "dataset": "scannet200", - "scene_name": "scene0252_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the plunger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0252_00_1", - "dataset": "scannet200", - "scene_name": "scene0252_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. table\nB. monitor\nC. desk\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "scene0252_00_2", - "dataset": "scannet200", - "scene_name": "scene0252_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the table, about 28 cm away." - }, - { - "id": "scene0252_00_3", - "dataset": "scannet200", - "scene_name": "scene0252_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 27.6}]}" - }, - { - "id": "scene0255_00_0", - "dataset": "scannet200", - "scene_name": "scene0255_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the candle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0255_00_1", - "dataset": "scannet200", - "scene_name": "scene0255_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. bucket\nB. printer\nC. cabinet\nD. paper cutter", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. paper cutter" - }, - { - "id": "scene0255_00_2", - "dataset": "scannet200", - "scene_name": "scene0255_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the counter. It is also to the side of and above the paper cutter, about 6 cm away. It is further to the side of and below the printer, about 25 cm away." - }, - { - "id": "scene0255_00_3", - "dataset": "scannet200", - "scene_name": "scene0255_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"counter\", \"references\": [{\"object\": \"paper cutter\", \"relation\": \"side_above\", \"distance_cm\": 5.7}, {\"object\": \"printer\", \"relation\": \"side_below\", \"distance_cm\": 25.0}]}" - }, - { - "id": "scene0255_00_4", - "dataset": "scannet200", - "scene_name": "scene0255_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the paper cutter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0255_00_5", - "dataset": "scannet200", - "scene_name": "scene0255_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper cutter in this scene?\nA. bottle\nB. cabinet\nC. printer\nD. bucket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. bottle" - }, - { - "id": "scene0255_00_6", - "dataset": "scannet200", - "scene_name": "scene0255_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper cutter in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper cutter is to the side of and below the bottle, about 6 cm away. It is also to the side of and above the cabinet, about 11 cm away. It is further to the side of and below the bucket, about 21 cm away." - }, - { - "id": "scene0255_00_7", - "dataset": "scannet200", - "scene_name": "scene0255_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper cutter in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper cutter\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 5.7}, {\"object\": \"cabinet\", \"relation\": \"side_above\", \"distance_cm\": 10.6}]}" - }, - { - "id": "scene0262_00_0", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0262_00_1", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. bottle\nB. clock\nC. pillow\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. pillow" - }, - { - "id": "scene0262_00_2", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the pillow, about 22 cm away. It is also to the side of and below the book, about 71 cm away." - }, - { - "id": "scene0262_00_3", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"couch\", \"references\": [{\"object\": \"pillow\", \"relation\": \"side_below\", \"distance_cm\": 21.7}, {\"object\": \"book\", \"relation\": \"side_below\", \"distance_cm\": 70.8}]}" - }, - { - "id": "scene0262_00_4", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the hair dryer appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0262_00_5", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the book in this scene?\nA. plant\nB. clock\nC. bottle\nD. backpack", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. plant" - }, - { - "id": "scene0262_00_6", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the book in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The book is to the side of and above the plant, about 27 cm away. It is also to the side of and above the backpack, about 71 cm away." - }, - { - "id": "scene0262_00_7", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"backpack\", \"references\": [{\"object\": \"plant\", \"relation\": \"side_above\", \"distance_cm\": 26.6}, {\"object\": \"backpack\", \"relation\": \"side_above\", \"distance_cm\": 70.8}]}" - }, - { - "id": "scene0262_00_8", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the water pitcher appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0262_00_9", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. office chair\nB. backpack\nC. telephone\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. telephone" - }, - { - "id": "scene0262_00_10", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the telephone, about 18 cm away. It is also to the side of and above the office chair, about 54 cm away." - }, - { - "id": "scene0262_00_11", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"office chair\", \"references\": [{\"object\": \"telephone\", \"relation\": \"side_above\", \"distance_cm\": 18.0}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 54.4}]}" - }, - { - "id": "scene0262_00_12", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the broom appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0262_00_13", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "13", - "question_type": "target presence verification", - "question": "In the entire scene, did the telephone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0262_00_14", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "14", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the telephone in this scene?\nA. bottle\nB. backpack\nC. book\nD. office chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. bottle" - }, - { - "id": "scene0262_00_15", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "15", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the telephone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The telephone is on the desk. It is also to the side of and below the bottle, about 18 cm away. It is further to the side of and above the office chair, about 56 cm away." - }, - { - "id": "scene0262_00_16", - "dataset": "scannet200", - "scene_name": "scene0262_00", - "sample_id": "16", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the telephone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"telephone\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 18.0}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 55.5}]}" - }, - { - "id": "scene0264_00_0", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bowl appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0264_00_1", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the book in this scene?\nA. board\nB. bulletin board\nC. monitor\nD. books", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. monitor" - }, - { - "id": "scene0264_00_2", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the book in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The book is to the side of and below the monitor, about 18 cm away." - }, - { - "id": "scene0264_00_3", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"books\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 18.5}]}" - }, - { - "id": "scene0264_00_4", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0264_00_5", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. printer\nB. computer tower\nC. monitor\nD. bulletin board", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. monitor" - }, - { - "id": "scene0264_00_6", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is to the side of and below the monitor, about 12 cm away. It is also to the side of and below the bulletin board, about 38 cm away. It is further to the side of and above the computer tower, about 43 cm away." - }, - { - "id": "scene0264_00_7", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"office chair\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 12.2}, {\"object\": \"bulletin board\", \"relation\": \"side_below\", \"distance_cm\": 37.9}]}" - }, - { - "id": "scene0264_00_8", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the tray appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0264_00_9", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tray in this scene?\nA. monitor\nB. office chair\nC. picture\nD. bulletin board", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. monitor" - }, - { - "id": "scene0264_00_10", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tray in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tray is to the side of and below the monitor, about 26 cm away. It is also to the side of and above the office chair, about 41 cm away. It is further to the side of and below the bulletin board, about 42 cm away." - }, - { - "id": "scene0264_00_11", - "dataset": "scannet200", - "scene_name": "scene0264_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tray in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tray\", \"support_surface\": \"office chair\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 25.7}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 41.4}]}" - }, - { - "id": "scene0274_00_0", - "dataset": "scannet200", - "scene_name": "scene0274_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the alarm clock appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0274_00_1", - "dataset": "scannet200", - "scene_name": "scene0274_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the telephone in this scene?\nA. table\nB. monitor\nC. box\nD. recycling bin", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. box" - }, - { - "id": "scene0274_00_2", - "dataset": "scannet200", - "scene_name": "scene0274_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the telephone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The telephone is on the table. It is also to the side of and below the box, about 17 cm away. It is further to the side of and below the monitor, about 31 cm away." - }, - { - "id": "scene0274_00_3", - "dataset": "scannet200", - "scene_name": "scene0274_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the telephone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"telephone\", \"support_surface\": \"table\", \"references\": [{\"object\": \"box\", \"relation\": \"side_below\", \"distance_cm\": 17.5}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 30.8}]}" - }, - { - "id": "scene0278_00_0", - "dataset": "scannet200", - "scene_name": "scene0278_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the broom appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0278_00_1", - "dataset": "scannet200", - "scene_name": "scene0278_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the broom in this scene?\nA. counter\nB. cabinets\nC. trash can\nD. copier", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. copier" - }, - { - "id": "scene0278_00_2", - "dataset": "scannet200", - "scene_name": "scene0278_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the broom in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The broom is to the side of and below the copier, about 42 cm away. It is also to the side of and above the trash can, about 54 cm away." - }, - { - "id": "scene0278_00_3", - "dataset": "scannet200", - "scene_name": "scene0278_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the broom in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"broom\", \"support_surface\": \"trash can\", \"references\": [{\"object\": \"copier\", \"relation\": \"side_below\", \"distance_cm\": 42.1}, {\"object\": \"trash can\", \"relation\": \"side_above\", \"distance_cm\": 53.8}]}" - }, - { - "id": "scene0278_00_4", - "dataset": "scannet200", - "scene_name": "scene0278_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the tray appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0278_00_5", - "dataset": "scannet200", - "scene_name": "scene0278_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper cutter in this scene?\nA. broom\nB. copier\nC. counter\nD. cabinets", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. cabinets" - }, - { - "id": "scene0278_00_6", - "dataset": "scannet200", - "scene_name": "scene0278_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper cutter in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper cutter is on the counter. It is also to the side of and above the cabinets, about 8 cm away." - }, - { - "id": "scene0278_00_7", - "dataset": "scannet200", - "scene_name": "scene0278_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper cutter in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper cutter\", \"support_surface\": \"counter\", \"references\": [{\"object\": \"cabinets\", \"relation\": \"side_above\", \"distance_cm\": 8.5}]}" - }, - { - "id": "scene0288_00_0", - "dataset": "scannet200", - "scene_name": "scene0288_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the book appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0288_00_1", - "dataset": "scannet200", - "scene_name": "scene0288_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the telephone in this scene?\nA. chair\nB. desk\nC. monitor\nD. backpack", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "scene0288_00_2", - "dataset": "scannet200", - "scene_name": "scene0288_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the telephone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The telephone is to the side of and above the chair, about 37 cm away." - }, - { - "id": "scene0288_00_3", - "dataset": "scannet200", - "scene_name": "scene0288_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the telephone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"telephone\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 37.5}]}" - }, - { - "id": "scene0304_00_0", - "dataset": "scannet200", - "scene_name": "scene0304_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0304_00_1", - "dataset": "scannet200", - "scene_name": "scene0304_00", - "sample_id": "1", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"cabinet\", \"references\": []}" - }, - { - "id": "scene0310_00_0", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the plate appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0310_00_1", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the plate in this scene?\nA. box\nB. dish rack\nC. coffee maker\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. box" - }, - { - "id": "scene0310_00_2", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the plate in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The plate is to the side of and above the box, about 23 cm away. It is also near the dish rack, about 28 cm away. It is further to the side of and above the cabinet, about 48 cm away." - }, - { - "id": "scene0310_00_3", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the plate in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"plate\", \"support_surface\": \"dish rack\", \"references\": [{\"object\": \"box\", \"relation\": \"side_above\", \"distance_cm\": 22.9}, {\"object\": \"dish rack\", \"relation\": \"near\", \"distance_cm\": 27.6}]}" - }, - { - "id": "scene0310_00_4", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dispenser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0310_00_5", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dispenser in this scene?\nA. kitchen counter\nB. sink\nC. dish rack\nD. kitchen cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. kitchen cabinet" - }, - { - "id": "scene0310_00_6", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dispenser is to the side of and below the kitchen cabinet, about 24 cm away. It is also to the side of and above the kitchen counter, about 32 cm away. It is further to the side of and above the dish rack, about 34 cm away." - }, - { - "id": "scene0310_00_7", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dispenser\", \"support_surface\": \"dish rack\", \"references\": [{\"object\": \"kitchen cabinet\", \"relation\": \"side_below\", \"distance_cm\": 24.3}, {\"object\": \"kitchen counter\", \"relation\": \"side_above\", \"distance_cm\": 32.4}]}" - }, - { - "id": "scene0310_00_8", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue box appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0310_00_9", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tray in this scene?\nA. cabinet\nB. shelf\nC. coffee maker\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. coffee maker" - }, - { - "id": "scene0310_00_10", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tray in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tray is to the side of and below the coffee maker, about 26 cm away. It is also to the side of and below the shelf, about 48 cm away." - }, - { - "id": "scene0310_00_11", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tray in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tray\", \"support_surface\": \"kitchen cabinet\", \"references\": [{\"object\": \"coffee maker\", \"relation\": \"side_below\", \"distance_cm\": 26.2}, {\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 48.3}]}" - }, - { - "id": "scene0310_00_12", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0310_00_13", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the water pitcher in this scene?\nA. soap dispenser\nB. dish rack\nC. cabinet\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. box" - }, - { - "id": "scene0310_00_14", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water pitcher in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water pitcher is to the side of and above the box, about 26 cm away. It is also to the side of and above the dish rack, about 29 cm away. It is further to the side of and above the cabinet, about 51 cm away." - }, - { - "id": "scene0310_00_15", - "dataset": "scannet200", - "scene_name": "scene0310_00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water pitcher in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water pitcher\", \"support_surface\": \"dish rack\", \"references\": [{\"object\": \"box\", \"relation\": \"side_above\", \"distance_cm\": 25.9}, {\"object\": \"dish rack\", \"relation\": \"side_above\", \"distance_cm\": 28.6}]}" - }, - { - "id": "scene0327_00_0", - "dataset": "scannet200", - "scene_name": "scene0327_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue box appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0327_00_1", - "dataset": "scannet200", - "scene_name": "scene0327_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the paper towel roll in this scene?\nA. sink\nB. kitchen cabinet\nC. range hood\nD. stove", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. stove" - }, - { - "id": "scene0327_00_2", - "dataset": "scannet200", - "scene_name": "scene0327_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the paper towel roll in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The paper towel roll is to the side of and above the stove, about 10 cm away. It is also to the side of and below the kitchen cabinet, about 16 cm away. It is further to the side of and below the range hood, about 38 cm away." - }, - { - "id": "scene0327_00_3", - "dataset": "scannet200", - "scene_name": "scene0327_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the paper towel roll in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"paper towel roll\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"stove\", \"relation\": \"side_above\", \"distance_cm\": 10.1}, {\"object\": \"kitchen cabinet\", \"relation\": \"side_below\", \"distance_cm\": 16.5}]}" - }, - { - "id": "scene0327_00_4", - "dataset": "scannet200", - "scene_name": "scene0327_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the telephone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0331_00_0", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0331_00_1", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. dresser\nB. recycling bin\nC. trash can\nD. shoes", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. shoes" - }, - { - "id": "scene0331_00_2", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the shoes, about 3 cm away. It is also to the side of and below the trash can, about 11 cm away. It is further to the side of and below the recycling bin, about 31 cm away." - }, - { - "id": "scene0331_00_3", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"shoes\", \"relation\": \"side_above\", \"distance_cm\": 2.6}, {\"object\": \"trash can\", \"relation\": \"side_below\", \"distance_cm\": 10.8}]}" - }, - { - "id": "scene0331_00_4", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the shoe appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0331_00_5", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bag in this scene?\nA. plant\nB. cup\nC. clothes\nD. bookshelf", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bookshelf" - }, - { - "id": "scene0331_00_6", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bag is to the side of and above the bookshelf, about 5 cm away. It is also to the side of and above the cup, about 14 cm away. It is further to the side of and below the plant, about 16 cm away." - }, - { - "id": "scene0331_00_7", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bag\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"bookshelf\", \"relation\": \"side_above\", \"distance_cm\": 5.2}, {\"object\": \"cup\", \"relation\": \"side_above\", \"distance_cm\": 13.8}]}" - }, - { - "id": "scene0331_00_8", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the book appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0331_00_9", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the book in this scene?\nA. fan\nB. plant\nC. cup\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. cup" - }, - { - "id": "scene0331_00_10", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the book in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The book is on the bookshelf. It is also to the side of and above the cup, about 9 cm away. It is further to the side of and above the fan, about 12 cm away." - }, - { - "id": "scene0331_00_11", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"cup\", \"relation\": \"side_above\", \"distance_cm\": 9.2}, {\"object\": \"fan\", \"relation\": \"side_above\", \"distance_cm\": 12.5}]}" - }, - { - "id": "scene0331_00_12", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the ball appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0331_00_13", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the purse in this scene?\nA. chair\nB. box\nC. block\nD. storage organizer", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. storage organizer" - }, - { - "id": "scene0331_00_14", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the purse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The purse is to the side of and below the storage organizer, about 9 cm away. It is also to the side of and below the box, about 11 cm away. It is further next to the block, about 11 cm away." - }, - { - "id": "scene0331_00_15", - "dataset": "scannet200", - "scene_name": "scene0331_00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the purse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"purse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"storage organizer\", \"relation\": \"side_below\", \"distance_cm\": 9.4}, {\"object\": \"box\", \"relation\": \"side_below\", \"distance_cm\": 10.6}]}" - }, - { - "id": "scene0333_00_0", - "dataset": "scannet200", - "scene_name": "scene0333_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the toaster appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0333_00_1", - "dataset": "scannet200", - "scene_name": "scene0333_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the laptop in this scene?\nA. computer tower\nB. chair\nC. person\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. monitor" - }, - { - "id": "scene0333_00_2", - "dataset": "scannet200", - "scene_name": "scene0333_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the laptop in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The laptop is on the desk. It is also under the monitor, about 3 cm away. It is further to the side of and above the person, about 11 cm away." - }, - { - "id": "scene0333_00_3", - "dataset": "scannet200", - "scene_name": "scene0333_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the laptop in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"laptop\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"monitor\", \"relation\": \"under\", \"distance_cm\": 3.1}, {\"object\": \"person\", \"relation\": \"side_above\", \"distance_cm\": 11.2}]}" - }, - { - "id": "scene0336_00_0", - "dataset": "scannet200", - "scene_name": "scene0336_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the book appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0336_00_1", - "dataset": "scannet200", - "scene_name": "scene0336_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the book in this scene?\nA. chair\nB. books\nC. step stool\nD. picture", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. step stool" - }, - { - "id": "scene0336_00_2", - "dataset": "scannet200", - "scene_name": "scene0336_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the book in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The book is to the side of and above the step stool, about 16 cm away. It is also to the side of and below the picture, about 62 cm away." - }, - { - "id": "scene0336_00_3", - "dataset": "scannet200", - "scene_name": "scene0336_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"step stool\", \"references\": [{\"object\": \"step stool\", \"relation\": \"side_above\", \"distance_cm\": 16.1}, {\"object\": \"picture\", \"relation\": \"side_below\", \"distance_cm\": 61.9}]}" - }, - { - "id": "scene0336_00_4", - "dataset": "scannet200", - "scene_name": "scene0336_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0336_00_5", - "dataset": "scannet200", - "scene_name": "scene0336_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. book\nB. monitor\nC. desk\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. desk" - }, - { - "id": "scene0336_00_6", - "dataset": "scannet200", - "scene_name": "scene0336_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is on the desk. It is also to the side of and above the chair, about 28 cm away. It is further to the side of and below the monitor, about 39 cm away." - }, - { - "id": "scene0336_00_7", - "dataset": "scannet200", - "scene_name": "scene0336_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 28.2}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 39.4}]}" - }, - { - "id": "scene0342_00_0", - "dataset": "scannet200", - "scene_name": "scene0342_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the hair dryer appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0342_00_1", - "dataset": "scannet200", - "scene_name": "scene0342_00", - "sample_id": "1", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": []}" - }, - { - "id": "scene0347_00_0", - "dataset": "scannet200", - "scene_name": "scene0347_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the guitar appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0347_00_1", - "dataset": "scannet200", - "scene_name": "scene0347_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the coffee kettle in this scene?\nA. monitor\nB. kitchen counter\nC. coffee maker\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. coffee maker" - }, - { - "id": "scene0347_00_2", - "dataset": "scannet200", - "scene_name": "scene0347_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the coffee kettle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The coffee kettle is to the side of and below the coffee maker, about 15 cm away. It is also to the side of and above the cup, about 47 cm away." - }, - { - "id": "scene0347_00_3", - "dataset": "scannet200", - "scene_name": "scene0347_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the coffee kettle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"coffee kettle\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"coffee maker\", \"relation\": \"side_below\", \"distance_cm\": 14.8}, {\"object\": \"cup\", \"relation\": \"side_above\", \"distance_cm\": 47.1}]}" - }, - { - "id": "scene0349_00_0", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0349_00_1", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "1", - "question_type": "target presence verification", - "question": "In the entire scene, did the stuffed animal appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0349_00_2", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "2", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the broom in this scene?\nA. clothes dryer\nB. bag\nC. sink\nD. washing machine", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. sink" - }, - { - "id": "scene0349_00_3", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "3", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the broom in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The broom is to the side of and below the sink, about 13 cm away. It is also to the side of and below the washing machine, about 39 cm away." - }, - { - "id": "scene0349_00_4", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "4", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the broom in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"broom\", \"support_surface\": \"dustpan\", \"references\": [{\"object\": \"sink\", \"relation\": \"side_below\", \"distance_cm\": 13.2}, {\"object\": \"washing machine\", \"relation\": \"side_below\", \"distance_cm\": 38.7}]}" - }, - { - "id": "scene0349_00_5", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "5", - "question_type": "target presence verification", - "question": "In the entire scene, did the shoe appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0349_00_6", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "6", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the dustpan in this scene?\nA. washing machine\nB. sink\nC. bag\nD. trash can", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. trash can" - }, - { - "id": "scene0349_00_7", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the dustpan in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The dustpan is next to the trash can, about 6 cm away. It is also to the side of and below the sink, about 12 cm away. It is further to the side of and below the washing machine, about 42 cm away." - }, - { - "id": "scene0349_00_8", - "dataset": "scannet200", - "scene_name": "scene0349_00", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the dustpan in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"dustpan\", \"support_surface\": \"broom\", \"references\": [{\"object\": \"trash can\", \"relation\": \"next_to\", \"distance_cm\": 5.6}, {\"object\": \"sink\", \"relation\": \"side_below\", \"distance_cm\": 12.5}]}" - }, - { - "id": "scene0352_00_0", - "dataset": "scannet200", - "scene_name": "scene0352_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0352_00_1", - "dataset": "scannet200", - "scene_name": "scene0352_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. computer tower\nB. shelf\nC. cabinet\nD. pillow", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. shelf" - }, - { - "id": "scene0352_00_2", - "dataset": "scannet200", - "scene_name": "scene0352_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the couch. It is also to the side of and below the shelf, about 12 cm away. It is further to the side of and above the pillow, about 46 cm away." - }, - { - "id": "scene0352_00_3", - "dataset": "scannet200", - "scene_name": "scene0352_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"couch\", \"references\": [{\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 11.7}, {\"object\": \"pillow\", \"relation\": \"side_above\", \"distance_cm\": 45.6}]}" - }, - { - "id": "scene0352_00_4", - "dataset": "scannet200", - "scene_name": "scene0352_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the dumbbell appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0352_00_5", - "dataset": "scannet200", - "scene_name": "scene0352_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the telephone in this scene?\nA. backpack\nB. cabinet\nC. shelf\nD. computer tower", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. shelf" - }, - { - "id": "scene0352_00_6", - "dataset": "scannet200", - "scene_name": "scene0352_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the telephone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The telephone is on the cabinet. It is also to the side of and above the shelf, about 24 cm away. It is further to the side of and above the computer tower, about 41 cm away." - }, - { - "id": "scene0352_00_7", - "dataset": "scannet200", - "scene_name": "scene0352_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the telephone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"telephone\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"shelf\", \"relation\": \"side_above\", \"distance_cm\": 23.8}, {\"object\": \"computer tower\", \"relation\": \"side_above\", \"distance_cm\": 40.6}]}" - }, - { - "id": "scene0353_00_0", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0353_00_1", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. chair\nB. blanket\nC. shelf\nD. laptop", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. laptop" - }, - { - "id": "scene0353_00_2", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the desk. It is also to the side of and above the laptop, about 20 cm away. It is further to the side of and above the chair, about 40 cm away." - }, - { - "id": "scene0353_00_3", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"laptop\", \"relation\": \"side_above\", \"distance_cm\": 20.5}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 39.9}]}" - }, - { - "id": "scene0353_00_4", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the bowl appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0353_00_5", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bowl in this scene?\nA. jar\nB. book\nC. laptop\nD. shelf", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. jar" - }, - { - "id": "scene0353_00_6", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bowl in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bowl is on the desk. It is also under the jar, about 3 cm away. It is further next to the book, about 13 cm away." - }, - { - "id": "scene0353_00_7", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bowl in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bowl\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"jar\", \"relation\": \"under\", \"distance_cm\": 3.3}, {\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 12.7}]}" - }, - { - "id": "scene0353_00_8", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the paper bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0353_00_9", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the laptop in this scene?\nA. desk lamp\nB. chair\nC. book\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. desk lamp" - }, - { - "id": "scene0353_00_10", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the laptop in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The laptop is to the side of and below the desk lamp, about 19 cm away. It is also to the side of and below the bottle, about 20 cm away. It is further to the side of and above the chair, about 24 cm away." - }, - { - "id": "scene0353_00_11", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the laptop in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"laptop\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"desk lamp\", \"relation\": \"side_below\", \"distance_cm\": 19.2}, {\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 20.5}]}" - }, - { - "id": "scene0353_00_12", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the dustpan appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0353_00_13", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the plate in this scene?\nA. book\nB. backpack\nC. shelf\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. book" - }, - { - "id": "scene0353_00_14", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the plate in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The plate is near the book, about 31 cm away. It is also to the side of and above the chair, about 34 cm away. It is further to the side of and below the shelf, about 43 cm away." - }, - { - "id": "scene0353_00_15", - "dataset": "scannet200", - "scene_name": "scene0353_00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the plate in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"plate\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"book\", \"relation\": \"near\", \"distance_cm\": 31.4}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 34.2}]}" - }, - { - "id": "scene0358_00_0", - "dataset": "scannet200", - "scene_name": "scene0358_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the alarm clock appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0358_00_1", - "dataset": "scannet200", - "scene_name": "scene0358_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. desk\nB. monitor\nC. chair\nD. keyboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. chair" - }, - { - "id": "scene0369_00_0", - "dataset": "scannet200", - "scene_name": "scene0369_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0370_00_0", - "dataset": "scannet200", - "scene_name": "scene0370_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the book appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0370_00_1", - "dataset": "scannet200", - "scene_name": "scene0370_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. stand\nB. office chair\nC. recycling bin\nD. microwave", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. stand" - }, - { - "id": "scene0370_00_2", - "dataset": "scannet200", - "scene_name": "scene0370_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also under the stand, about 5 cm away. It is further to the side of and above the office chair, about 7 cm away." - }, - { - "id": "scene0370_00_3", - "dataset": "scannet200", - "scene_name": "scene0370_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"stand\", \"relation\": \"under\", \"distance_cm\": 5.0}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 7.1}]}" - }, - { - "id": "scene0370_00_4", - "dataset": "scannet200", - "scene_name": "scene0370_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0370_00_5", - "dataset": "scannet200", - "scene_name": "scene0370_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the water bottle in this scene?\nA. table\nB. trash can\nC. recycling bin\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "scene0370_00_6", - "dataset": "scannet200", - "scene_name": "scene0370_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water bottle is to the side of and below the table, about 5 cm away. It is also to the side of and below the recycling bin, about 14 cm away. It is further to the side of and below the trash can, about 50 cm away." - }, - { - "id": "scene0370_00_7", - "dataset": "scannet200", - "scene_name": "scene0370_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 5.0}, {\"object\": \"recycling bin\", \"relation\": \"side_below\", \"distance_cm\": 14.0}]}" - }, - { - "id": "scene0404_00_0", - "dataset": "scannet200", - "scene_name": "scene0404_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0404_00_1", - "dataset": "scannet200", - "scene_name": "scene0404_00", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also to the side of and above the swiffer, about 44 cm away. It is further to the side of and above the dustpan, about 44 cm away." - }, - { - "id": "scene0404_00_2", - "dataset": "scannet200", - "scene_name": "scene0404_00", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"swiffer\", \"relation\": \"side_above\", \"distance_cm\": 44.4}, {\"object\": \"dustpan\", \"relation\": \"side_above\", \"distance_cm\": 44.4}]}" - }, - { - "id": "scene0404_00_3", - "dataset": "scannet200", - "scene_name": "scene0404_00", - "sample_id": "3", - "question_type": "target presence verification", - "question": "In the entire scene, did the cd case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0404_00_4", - "dataset": "scannet200", - "scene_name": "scene0404_00", - "sample_id": "4", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the dustpan in this scene?\nA. bottle\nB. storage bin\nC. table\nD. cloth", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. table" - }, - { - "id": "scene0404_00_5", - "dataset": "scannet200", - "scene_name": "scene0404_00", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the dustpan in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The dustpan is under the table, about 3 cm away. It is also to the side of and above the storage bin, about 28 cm away. It is further to the side of and below the bottle, about 44 cm away." - }, - { - "id": "scene0404_00_6", - "dataset": "scannet200", - "scene_name": "scene0404_00", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the dustpan in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"dustpan\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"under\", \"distance_cm\": 3.1}, {\"object\": \"storage bin\", \"relation\": \"side_above\", \"distance_cm\": 28.0}]}" - }, - { - "id": "scene0409_00_0", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the hat appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0409_00_1", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the coffee kettle in this scene?\nA. microwave\nB. dishwasher\nC. kitchen cabinets\nD. coffee maker", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. microwave" - }, - { - "id": "scene0409_00_2", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the coffee kettle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The coffee kettle is next to the microwave, about 7 cm away. It is also to the side of and above the kitchen cabinets, about 26 cm away. It is further to the side of and above the dishwasher, about 39 cm away." - }, - { - "id": "scene0409_00_3", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the coffee kettle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"coffee kettle\", \"support_surface\": \"coffee maker\", \"references\": [{\"object\": \"microwave\", \"relation\": \"next_to\", \"distance_cm\": 6.7}, {\"object\": \"kitchen cabinets\", \"relation\": \"side_above\", \"distance_cm\": 25.7}]}" - }, - { - "id": "scene0409_00_4", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the fire extinguisher appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0409_00_5", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the fire extinguisher in this scene?\nA. chair\nB. dishwasher\nC. coffee maker\nD. coffee kettle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "scene0409_00_6", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the fire extinguisher in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The fire extinguisher is to the side of and below the chair, about 55 cm away." - }, - { - "id": "scene0409_00_7", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the fire extinguisher in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"fire extinguisher\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 54.6}]}" - }, - { - "id": "scene0409_00_8", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the stuffed animal appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0409_00_9", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the plate in this scene?\nA. coffee maker\nB. soap dispenser\nC. dishwasher\nD. kitchen cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. coffee maker" - }, - { - "id": "scene0409_00_10", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the plate in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The plate is to the side of and below the coffee maker, about 23 cm away. It is also to the side of and above the kitchen cabinet, about 43 cm away. It is further to the side of and above the dishwasher, about 43 cm away." - }, - { - "id": "scene0409_00_11", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the plate in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"plate\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"coffee maker\", \"relation\": \"side_below\", \"distance_cm\": 22.8}, {\"object\": \"kitchen cabinet\", \"relation\": \"side_above\", \"distance_cm\": 42.6}]}" - }, - { - "id": "scene0409_00_12", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dispenser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0409_00_13", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dispenser in this scene?\nA. chair\nB. kitchen counter\nC. plate\nD. kitchen cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. kitchen counter" - }, - { - "id": "scene0409_00_14", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dispenser is to the side of and above the kitchen counter, about 13 cm away. It is also to the side of and above the plate, about 61 cm away. It is further to the side of and above the kitchen cabinet, about 66 cm away." - }, - { - "id": "scene0409_00_15", - "dataset": "scannet200", - "scene_name": "scene0409_00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dispenser\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"kitchen counter\", \"relation\": \"side_above\", \"distance_cm\": 13.4}, {\"object\": \"plate\", \"relation\": \"side_above\", \"distance_cm\": 60.7}]}" - }, - { - "id": "scene0430_00_0", - "dataset": "scannet200", - "scene_name": "scene0430_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the guitar appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0434_00_0", - "dataset": "scannet200", - "scene_name": "scene0434_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0434_00_1", - "dataset": "scannet200", - "scene_name": "scene0434_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dish in this scene?\nA. ledge\nB. bathroom vanity\nC. towel\nD. bathtub", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bathtub" - }, - { - "id": "scene0434_00_2", - "dataset": "scannet200", - "scene_name": "scene0434_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dish in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dish is to the side of and above the bathtub, about 26 cm away. It is also to the side of and above the towel, about 62 cm away." - }, - { - "id": "scene0434_00_3", - "dataset": "scannet200", - "scene_name": "scene0434_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dish in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dish\", \"support_surface\": \"shower walls\", \"references\": [{\"object\": \"bathtub\", \"relation\": \"side_above\", \"distance_cm\": 26.2}, {\"object\": \"towel\", \"relation\": \"side_above\", \"distance_cm\": 61.5}]}" - }, - { - "id": "scene0438_00_0", - "dataset": "scannet200", - "scene_name": "scene0438_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the fire extinguisher appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0438_00_1", - "dataset": "scannet200", - "scene_name": "scene0438_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bag in this scene?\nA. nightstand\nB. tissue box\nC. dresser\nD. ball", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. ball" - }, - { - "id": "scene0438_00_2", - "dataset": "scannet200", - "scene_name": "scene0438_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bag is next to the ball. It is also to the side of and below the dresser, about 35 cm away. It is further to the side of and below the nightstand, about 39 cm away." - }, - { - "id": "scene0438_00_3", - "dataset": "scannet200", - "scene_name": "scene0438_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bag\", \"support_surface\": \"dresser\", \"references\": [{\"object\": \"ball\", \"relation\": \"next_to\", \"distance_cm\": 2.4}, {\"object\": \"dresser\", \"relation\": \"side_below\", \"distance_cm\": 35.3}]}" - }, - { - "id": "scene0446_00_0", - "dataset": "scannet200", - "scene_name": "scene0446_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the water pitcher appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0446_00_1", - "dataset": "scannet200", - "scene_name": "scene0446_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toilet paper in this scene?\nA. radiator\nB. rod\nC. towel\nD. mat", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. towel" - }, - { - "id": "scene0446_00_2", - "dataset": "scannet200", - "scene_name": "scene0446_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toilet paper in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toilet paper is to the side of and below the towel, about 25 cm away. It is also to the side of and above the radiator, about 29 cm away. It is further to the side of and above the mat, about 44 cm away." - }, - { - "id": "scene0446_00_3", - "dataset": "scannet200", - "scene_name": "scene0446_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toilet paper in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toilet paper\", \"support_surface\": \"radiator\", \"references\": [{\"object\": \"towel\", \"relation\": \"side_below\", \"distance_cm\": 24.8}, {\"object\": \"radiator\", \"relation\": \"side_above\", \"distance_cm\": 29.3}]}" - }, - { - "id": "scene0476_00_0", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the guitar appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0476_00_1", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. monitor\nB. office chair\nC. keyboard\nD. file cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. file cabinet" - }, - { - "id": "scene0476_00_2", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the file cabinet, about 17 cm away. It is also to the side of and below the monitor, about 39 cm away. It is further to the side of and above the office chair, about 44 cm away." - }, - { - "id": "scene0476_00_3", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"file cabinet\", \"references\": [{\"object\": \"file cabinet\", \"relation\": \"side_above\", \"distance_cm\": 17.1}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 39.2}]}" - }, - { - "id": "scene0476_00_4", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0476_00_5", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the book in this scene?\nA. monitor\nB. clothing\nC. file cabinet\nD. office chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. file cabinet" - }, - { - "id": "scene0476_00_6", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the book in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The book is to the side of and above the file cabinet, about 25 cm away. It is also to the side of and above the office chair, about 27 cm away. It is further to the side of and above the clothing, about 30 cm away." - }, - { - "id": "scene0476_00_7", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"file cabinet\", \"references\": [{\"object\": \"file cabinet\", \"relation\": \"side_above\", \"distance_cm\": 24.6}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 27.0}]}" - }, - { - "id": "scene0476_00_8", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0476_00_9", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. keyboard\nB. telephone\nC. monitor\nD. office chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. keyboard" - }, - { - "id": "scene0476_00_10", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also next to the keyboard, about 10 cm away. It is further near the telephone, about 31 cm away." - }, - { - "id": "scene0476_00_11", - "dataset": "scannet200", - "scene_name": "scene0476_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 9.6}, {\"object\": \"telephone\", \"relation\": \"near\", \"distance_cm\": 30.5}]}" - }, - { - "id": "scene0477_00_0", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0477_00_1", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. dresser\nB. shoes\nC. clothing\nD. lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. shoes" - }, - { - "id": "scene0477_00_2", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the shoes, about 5 cm away. It is also to the side of and below the clothing, about 36 cm away." - }, - { - "id": "scene0477_00_3", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"suitcase\", \"references\": [{\"object\": \"shoes\", \"relation\": \"side_above\", \"distance_cm\": 4.8}, {\"object\": \"clothing\", \"relation\": \"side_below\", \"distance_cm\": 36.4}]}" - }, - { - "id": "scene0477_00_4", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the telephone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0477_00_5", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the telephone in this scene?\nA. backpack\nB. pillow\nC. lamp\nD. clothing", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. pillow" - }, - { - "id": "scene0477_00_6", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the telephone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The telephone is on the nightstand. It is also to the side of and below the pillow, about 22 cm away. It is further to the side of and below the lamp, about 63 cm away." - }, - { - "id": "scene0477_00_7", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the telephone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"telephone\", \"support_surface\": \"nightstand\", \"references\": [{\"object\": \"pillow\", \"relation\": \"side_below\", \"distance_cm\": 21.9}, {\"object\": \"lamp\", \"relation\": \"side_below\", \"distance_cm\": 63.4}]}" - }, - { - "id": "scene0477_00_8", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the guitar appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0477_00_9", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toilet paper in this scene?\nA. picture\nB. towel\nC. backpack\nD. trash can", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. towel" - }, - { - "id": "scene0477_00_10", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toilet paper in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toilet paper is to the side of and below the towel, about 19 cm away. It is also to the side of and below the picture, about 36 cm away. It is further to the side of and above the trash can, about 57 cm away." - }, - { - "id": "scene0477_00_11", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toilet paper in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toilet paper\", \"support_surface\": \"toilet\", \"references\": [{\"object\": \"towel\", \"relation\": \"side_below\", \"distance_cm\": 19.2}, {\"object\": \"picture\", \"relation\": \"side_below\", \"distance_cm\": 36.0}]}" - }, - { - "id": "scene0477_00_12", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the hair dryer appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0477_00_13", - "dataset": "scannet200", - "scene_name": "scene0477_00", - "sample_id": "13", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tray in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tray\", \"support_surface\": \"dresser\", \"references\": []}" - }, - { - "id": "scene0508_00_0", - "dataset": "scannet200", - "scene_name": "scene0508_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the paper towel roll appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0508_00_1", - "dataset": "scannet200", - "scene_name": "scene0508_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. telephone\nB. chair\nC. thermostat\nD. trash can", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. trash can" - }, - { - "id": "scene0508_00_2", - "dataset": "scannet200", - "scene_name": "scene0508_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the trash can, about 39 cm away." - }, - { - "id": "scene0508_00_3", - "dataset": "scannet200", - "scene_name": "scene0508_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"trash can\", \"references\": [{\"object\": \"trash can\", \"relation\": \"side_below\", \"distance_cm\": 38.8}]}" - }, - { - "id": "scene0508_00_4", - "dataset": "scannet200", - "scene_name": "scene0508_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the telephone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0508_00_5", - "dataset": "scannet200", - "scene_name": "scene0508_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the telephone in this scene?\nA. thermostat\nB. trash can\nC. backpack\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. thermostat" - }, - { - "id": "scene0508_00_6", - "dataset": "scannet200", - "scene_name": "scene0508_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the telephone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The telephone is to the side of and below the thermostat, about 16 cm away. It is also to the side of and above the chair, about 50 cm away." - }, - { - "id": "scene0508_00_7", - "dataset": "scannet200", - "scene_name": "scene0508_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the telephone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"telephone\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"thermostat\", \"relation\": \"side_below\", \"distance_cm\": 15.6}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 50.2}]}" - }, - { - "id": "scene0509_00_0", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the shoe appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0509_00_1", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bag in this scene?\nA. bathroom vanity\nB. sink\nC. mirror\nD. hair dryer", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. mirror" - }, - { - "id": "scene0509_00_2", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bag is under the mirror. It is also to the side of and above the sink, about 6 cm away." - }, - { - "id": "scene0509_00_3", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bag\", \"support_surface\": \"sink\", \"references\": [{\"object\": \"mirror\", \"relation\": \"under\", \"distance_cm\": 2.5}, {\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 6.0}]}" - }, - { - "id": "scene0509_00_4", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the hair dryer appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0509_00_5", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the hair dryer in this scene?\nA. towel\nB. toilet paper\nC. tissue box\nD. sink", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. towel" - }, - { - "id": "scene0509_00_6", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the hair dryer in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The hair dryer is to the side of and above the towel. It is also to the side of and above the toilet paper, about 27 cm away. It is further to the side of and below the sink, about 34 cm away." - }, - { - "id": "scene0509_00_7", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the hair dryer in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"hair dryer\", \"support_surface\": \"toilet paper\", \"references\": [{\"object\": \"towel\", \"relation\": \"side_above\", \"distance_cm\": 2.5}, {\"object\": \"toilet paper\", \"relation\": \"side_above\", \"distance_cm\": 27.5}]}" - }, - { - "id": "scene0509_00_8", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the broom appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0509_00_9", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. toilet paper\nB. sink\nC. hair dryer\nD. towel", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. sink" - }, - { - "id": "scene0509_00_10", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is to the side of and above the sink, about 16 cm away. It is also to the side of and above the towel, about 30 cm away. It is further to the side of and above the hair dryer, about 34 cm away." - }, - { - "id": "scene0509_00_11", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"toilet paper\", \"references\": [{\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 15.5}, {\"object\": \"towel\", \"relation\": \"side_above\", \"distance_cm\": 30.1}]}" - }, - { - "id": "scene0509_00_12", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the toilet paper appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0509_00_13", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toilet paper in this scene?\nA. toilet\nB. towel\nC. bathroom vanity\nD. trash can", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. trash can" - }, - { - "id": "scene0509_00_14", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toilet paper in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toilet paper is above the trash can, about 8 cm away. It is also near the toilet, about 16 cm away. It is further to the side of and below the bathroom vanity, about 26 cm away." - }, - { - "id": "scene0509_00_15", - "dataset": "scannet200", - "scene_name": "scene0509_00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toilet paper in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toilet paper\", \"support_surface\": \"tissue box\", \"references\": [{\"object\": \"trash can\", \"relation\": \"above\", \"distance_cm\": 8.3}, {\"object\": \"toilet\", \"relation\": \"near\", \"distance_cm\": 15.9}]}" - }, - { - "id": "scene0518_00_0", - "dataset": "scannet200", - "scene_name": "scene0518_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dispenser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0518_00_1", - "dataset": "scannet200", - "scene_name": "scene0518_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. picture\nB. desk\nC. monitor\nD. office chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. desk" - }, - { - "id": "scene0518_00_2", - "dataset": "scannet200", - "scene_name": "scene0518_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the desk, about 5 cm away. It is also to the side of and below the picture, about 53 cm away." - }, - { - "id": "scene0518_00_3", - "dataset": "scannet200", - "scene_name": "scene0518_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"desk\", \"relation\": \"side_above\", \"distance_cm\": 4.6}, {\"object\": \"picture\", \"relation\": \"side_below\", \"distance_cm\": 53.0}]}" - }, - { - "id": "scene0522_00_0", - "dataset": "scannet200", - "scene_name": "scene0522_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the plunger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0522_00_1", - "dataset": "scannet200", - "scene_name": "scene0522_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the purse in this scene?\nA. clothes dryer\nB. cabinet\nC. basket\nD. bag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bag" - }, - { - "id": "scene0522_00_2", - "dataset": "scannet200", - "scene_name": "scene0522_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the purse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The purse is on the counter. It is also next to the bag, about 4 cm away. It is further to the side of and above the clothes dryer, about 8 cm away." - }, - { - "id": "scene0522_00_3", - "dataset": "scannet200", - "scene_name": "scene0522_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the purse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"purse\", \"support_surface\": \"counter\", \"references\": [{\"object\": \"bag\", \"relation\": \"next_to\", \"distance_cm\": 4.0}, {\"object\": \"clothes dryer\", \"relation\": \"side_above\", \"distance_cm\": 7.6}]}" - }, - { - "id": "scene0532_00_0", - "dataset": "scannet200", - "scene_name": "scene0532_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the hair dryer appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0532_00_1", - "dataset": "scannet200", - "scene_name": "scene0532_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the fire extinguisher in this scene?\nA. monitor\nB. chair\nC. desk\nD. cable", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. cable" - }, - { - "id": "scene0532_00_2", - "dataset": "scannet200", - "scene_name": "scene0532_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the fire extinguisher in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The fire extinguisher is to the side of and below the cable, about 81 cm away." - }, - { - "id": "scene0532_00_3", - "dataset": "scannet200", - "scene_name": "scene0532_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the fire extinguisher in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"fire extinguisher\", \"support_surface\": \"cable\", \"references\": [{\"object\": \"cable\", \"relation\": \"side_below\", \"distance_cm\": 80.8}]}" - }, - { - "id": "scene0535_00_0", - "dataset": "scannet200", - "scene_name": "scene0535_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the alarm clock appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0535_00_1", - "dataset": "scannet200", - "scene_name": "scene0535_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the book in this scene?\nA. backpack\nB. desk\nC. monitor\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "scene0535_00_2", - "dataset": "scannet200", - "scene_name": "scene0535_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the book in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The book is on the desk. It is also to the side of and above the chair, about 38 cm away." - }, - { - "id": "scene0535_00_3", - "dataset": "scannet200", - "scene_name": "scene0535_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the book in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"book\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 38.0}]}" - }, - { - "id": "scene0538_00_0", - "dataset": "scannet200", - "scene_name": "scene0538_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the plunger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0538_00_1", - "dataset": "scannet200", - "scene_name": "scene0538_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the plunger in this scene?\nA. toilet\nB. bathroom vanity\nC. mirror\nD. ledge", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. toilet" - }, - { - "id": "scene0538_00_2", - "dataset": "scannet200", - "scene_name": "scene0538_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the plunger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The plunger is near the toilet, about 20 cm away." - }, - { - "id": "scene0538_00_3", - "dataset": "scannet200", - "scene_name": "scene0538_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the plunger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"plunger\", \"support_surface\": \"toilet\", \"references\": [{\"object\": \"toilet\", \"relation\": \"near\", \"distance_cm\": 19.8}]}" - }, - { - "id": "scene0538_00_4", - "dataset": "scannet200", - "scene_name": "scene0538_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0538_00_5", - "dataset": "scannet200", - "scene_name": "scene0538_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dispenser in this scene?\nA. paper towel dispenser\nB. sink\nC. ledge\nD. bathroom vanity", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. ledge" - }, - { - "id": "scene0538_00_6", - "dataset": "scannet200", - "scene_name": "scene0538_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dispenser is to the side of and above the ledge, about 14 cm away. It is also to the side of and below the paper towel dispenser, about 22 cm away. It is further to the side of and above the bathroom vanity, about 30 cm away." - }, - { - "id": "scene0538_00_7", - "dataset": "scannet200", - "scene_name": "scene0538_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dispenser\", \"support_surface\": \"paper towel dispenser\", \"references\": [{\"object\": \"ledge\", \"relation\": \"side_above\", \"distance_cm\": 13.9}, {\"object\": \"paper towel dispenser\", \"relation\": \"side_below\", \"distance_cm\": 21.8}]}" - }, - { - "id": "scene0547_00_0", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0547_00_1", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. dresser\nB. laundry hamper\nC. bag\nD. computer tower", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. dresser" - }, - { - "id": "scene0547_00_2", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is near the dresser, about 16 cm away. It is also to the side of and above the laundry hamper, about 24 cm away. It is further to the side of and below the computer tower, about 34 cm away." - }, - { - "id": "scene0547_00_3", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"dresser\", \"references\": [{\"object\": \"dresser\", \"relation\": \"near\", \"distance_cm\": 15.8}, {\"object\": \"laundry hamper\", \"relation\": \"side_above\", \"distance_cm\": 23.9}]}" - }, - { - "id": "scene0547_00_4", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0547_00_5", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bag in this scene?\nA. clothing\nB. backpack\nC. clothes\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. clothing" - }, - { - "id": "scene0547_00_6", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bag is to the side of and below the clothing, about 32 cm away. It is also near the backpack, about 40 cm away. It is further to the side of and below the clothes, about 57 cm away." - }, - { - "id": "scene0547_00_7", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bag\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"clothing\", \"relation\": \"side_below\", \"distance_cm\": 32.1}, {\"object\": \"backpack\", \"relation\": \"near\", \"distance_cm\": 40.4}]}" - }, - { - "id": "scene0547_00_8", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the plate appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0547_00_9", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bowl in this scene?\nA. chair\nB. shelf\nC. mini fridge\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. shelf" - }, - { - "id": "scene0547_00_10", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bowl in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bowl is on the desk. It is also under the shelf, about 3 cm away. It is further to the side of and below the monitor, about 18 cm away." - }, - { - "id": "scene0547_00_11", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bowl in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bowl\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"shelf\", \"relation\": \"under\", \"distance_cm\": 2.5}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 18.1}]}" - }, - { - "id": "scene0547_00_12", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the water pitcher appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0547_00_13", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the water pitcher in this scene?\nA. shelf\nB. bowl\nC. chair\nD. pillow", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. shelf" - }, - { - "id": "scene0547_00_14", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water pitcher in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water pitcher is to the side of and below the shelf, about 32 cm away. It is also to the side of and above the bowl, about 45 cm away. It is further to the side of and above the pillow, about 45 cm away." - }, - { - "id": "scene0547_00_15", - "dataset": "scannet200", - "scene_name": "scene0547_00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water pitcher in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water pitcher\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 32.0}, {\"object\": \"bowl\", \"relation\": \"side_above\", \"distance_cm\": 45.1}]}" - }, - { - "id": "scene0563_00_0", - "dataset": "scannet200", - "scene_name": "scene0563_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the coffee kettle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0563_00_1", - "dataset": "scannet200", - "scene_name": "scene0563_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. keyboard\nB. monitor\nC. desk\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "scene0564_00_0", - "dataset": "scannet200", - "scene_name": "scene0564_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the toilet paper appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0564_00_1", - "dataset": "scannet200", - "scene_name": "scene0564_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the broom in this scene?\nA. sign\nB. sink\nC. trash can\nD. bathroom vanity", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bathroom vanity" - }, - { - "id": "scene0564_00_2", - "dataset": "scannet200", - "scene_name": "scene0564_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the broom in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The broom is to the side of and below the bathroom vanity, about 15 cm away. It is also to the side of and below the sink, about 38 cm away. It is further to the side of and above the trash can, about 54 cm away." - }, - { - "id": "scene0564_00_3", - "dataset": "scannet200", - "scene_name": "scene0564_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the broom in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"broom\", \"support_surface\": \"sign\", \"references\": [{\"object\": \"bathroom vanity\", \"relation\": \"side_below\", \"distance_cm\": 15.1}, {\"object\": \"sink\", \"relation\": \"side_below\", \"distance_cm\": 38.0}]}" - }, - { - "id": "scene0566_00_0", - "dataset": "scannet200", - "scene_name": "scene0566_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the purse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0566_00_1", - "dataset": "scannet200", - "scene_name": "scene0566_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. chair\nB. monitor\nC. table\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. table" - }, - { - "id": "scene0566_00_2", - "dataset": "scannet200", - "scene_name": "scene0566_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is near the table, about 17 cm away." - }, - { - "id": "scene0566_00_3", - "dataset": "scannet200", - "scene_name": "scene0566_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"near\", \"distance_cm\": 16.6}]}" - }, - { - "id": "scene0571_00_0", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0571_00_1", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "1", - "question_type": "target presence verification", - "question": "In the entire scene, did the book appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0571_00_2", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "2", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the plunger in this scene?\nA. toilet\nB. shampoo bottle\nC. backpack\nD. bathtub", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. toilet" - }, - { - "id": "scene0571_00_3", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "3", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the plunger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The plunger is to the side of and below the toilet, about 14 cm away. It is also to the side of and below the shampoo bottle, about 69 cm away." - }, - { - "id": "scene0571_00_4", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "4", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the plunger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"plunger\", \"support_surface\": \"bathtub\", \"references\": [{\"object\": \"toilet\", \"relation\": \"side_below\", \"distance_cm\": 13.7}, {\"object\": \"shampoo bottle\", \"relation\": \"side_below\", \"distance_cm\": 69.2}]}" - }, - { - "id": "scene0571_00_5", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "5", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0571_00_6", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "6", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap dish in this scene?\nA. bathtub\nB. backpack\nC. shampoo bottle\nD. conditioner bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. bathtub" - }, - { - "id": "scene0571_00_7", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap dish in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap dish is to the side of and above the bathtub, about 20 cm away. It is also to the side of and above the conditioner bottle, about 49 cm away. It is further to the side of and above the shampoo bottle, about 54 cm away." - }, - { - "id": "scene0571_00_8", - "dataset": "scannet200", - "scene_name": "scene0571_00", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap dish in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap dish\", \"support_surface\": \"shower walls\", \"references\": [{\"object\": \"bathtub\", \"relation\": \"side_above\", \"distance_cm\": 20.3}, {\"object\": \"conditioner bottle\", \"relation\": \"side_above\", \"distance_cm\": 49.3}]}" - }, - { - "id": "scene0580_00_0", - "dataset": "scannet200", - "scene_name": "scene0580_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0580_00_1", - "dataset": "scannet200", - "scene_name": "scene0580_00", - "sample_id": "1", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap dish appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0580_00_2", - "dataset": "scannet200", - "scene_name": "scene0580_00", - "sample_id": "2", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the hat in this scene?\nA. backpack\nB. bag\nC. bookshelf\nD. trash can", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. bookshelf" - }, - { - "id": "scene0580_00_3", - "dataset": "scannet200", - "scene_name": "scene0580_00", - "sample_id": "3", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the hat in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The hat is on the dresser. It is also to the side of and below the bookshelf, about 16 cm away. It is further to the side of and above the trash can, about 59 cm away." - }, - { - "id": "scene0580_00_4", - "dataset": "scannet200", - "scene_name": "scene0580_00", - "sample_id": "4", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the hat in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"hat\", \"support_surface\": \"dresser\", \"references\": [{\"object\": \"bookshelf\", \"relation\": \"side_below\", \"distance_cm\": 16.4}, {\"object\": \"trash can\", \"relation\": \"side_above\", \"distance_cm\": 59.2}]}" - }, - { - "id": "scene0582_00_0", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the alarm clock appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0582_00_1", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. telephone\nB. office chair\nC. monitor\nD. computer tower", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. office chair" - }, - { - "id": "scene0582_00_2", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also to the side of and above the office chair, about 12 cm away. It is further to the side of and below the monitor, about 18 cm away." - }, - { - "id": "scene0582_00_3", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 12.3}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 18.1}]}" - }, - { - "id": "scene0582_00_4", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "scene0582_00_5", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. keyboard\nB. telephone\nC. monitor\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. keyboard" - }, - { - "id": "scene0582_00_6", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also next to the keyboard. It is further to the side of and below the monitor, about 11 cm away." - }, - { - "id": "scene0582_00_7", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 2.2}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 10.7}]}" - }, - { - "id": "scene0582_00_8", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "scene0582_00_9", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. box\nB. book\nC. cabinet\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. desk" - }, - { - "id": "scene0582_00_10", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the desk, about 16 cm away. It is also to the side of and below the cabinet, about 53 cm away." - }, - { - "id": "scene0582_00_11", - "dataset": "scannet200", - "scene_name": "scene0582_00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"desk\", \"relation\": \"side_below\", \"distance_cm\": 15.9}, {\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 52.7}]}" - }, - { - "id": "00dd871005_0", - "dataset": "scannetpp", - "scene_name": "00dd871005", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the purse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "00dd871005_1", - "dataset": "scannetpp", - "scene_name": "00dd871005", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the hairbrush in this scene?\nA. pasta box\nB. rolling cabinet\nC. polaroid camera\nD. cardboard box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. cardboard box" - }, - { - "id": "00dd871005_2", - "dataset": "scannetpp", - "scene_name": "00dd871005", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the hairbrush in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The hairbrush is to the side of and below the cardboard box, about 7 cm away. It is also to the side of and above the rolling cabinet, about 17 cm away. It is further to the side of and above the polaroid camera, about 21 cm away." - }, - { - "id": "00dd871005_3", - "dataset": "scannetpp", - "scene_name": "00dd871005", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the hairbrush in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"hairbrush\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"cardboard box\", \"relation\": \"side_below\", \"distance_cm\": 6.6}, {\"object\": \"rolling cabinet\", \"relation\": \"side_above\", \"distance_cm\": 16.7}]}" - }, - { - "id": "00dd871005_4", - "dataset": "scannetpp", - "scene_name": "00dd871005", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "00dd871005_5", - "dataset": "scannetpp", - "scene_name": "00dd871005", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone charger in this scene?\nA. light switch\nB. pillow\nC. shelf\nD. projector", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. shelf" - }, - { - "id": "00dd871005_6", - "dataset": "scannetpp", - "scene_name": "00dd871005", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is near the shelf, about 28 cm away. It is also to the side of and below the projector, about 33 cm away. It is further above the light switch, about 50 cm away." - }, - { - "id": "00dd871005_7", - "dataset": "scannetpp", - "scene_name": "00dd871005", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"shelf\", \"relation\": \"near\", \"distance_cm\": 27.8}, {\"object\": \"projector\", \"relation\": \"side_below\", \"distance_cm\": 33.5}]}" - }, - { - "id": "2e74812d00_0", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mousepad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "2e74812d00_1", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cable in this scene?\nA. ball\nB. split\nC. computer tower\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. power socket" - }, - { - "id": "2e74812d00_2", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cable in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cable is under the power socket. It is also to the side of and above the computer tower, about 20 cm away. It is further to the side of and above the ball, about 22 cm away." - }, - { - "id": "2e74812d00_3", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cable in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cable\", \"support_surface\": \"office chair\", \"references\": [{\"object\": \"power socket\", \"relation\": \"under\", \"distance_cm\": 2.1}, {\"object\": \"computer tower\", \"relation\": \"side_above\", \"distance_cm\": 20.4}]}" - }, - { - "id": "2e74812d00_4", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "2e74812d00_5", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. pen\nB. plant pot\nC. plant\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. pen" - }, - { - "id": "2e74812d00_6", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also next to the pen, about 3 cm away. It is further next to the bottle, about 6 cm away." - }, - { - "id": "2e74812d00_7", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"pen\", \"relation\": \"next_to\", \"distance_cm\": 2.6}, {\"object\": \"bottle\", \"relation\": \"next_to\", \"distance_cm\": 5.7}]}" - }, - { - "id": "2e74812d00_8", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the power extension appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "2e74812d00_9", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the glasses case in this scene?\nA. office chair\nB. food container\nC. mug\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. monitor" - }, - { - "id": "2e74812d00_10", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the glasses case in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The glasses case is on the table. It is also under the monitor, about 4 cm away. It is further next to the mug, about 10 cm away." - }, - { - "id": "2e74812d00_11", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the glasses case in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"glasses case\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"under\", \"distance_cm\": 3.8}, {\"object\": \"mug\", \"relation\": \"next_to\", \"distance_cm\": 10.0}]}" - }, - { - "id": "2e74812d00_12", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the eraser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "2e74812d00_13", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphone case in this scene?\nA. table\nB. monitor\nC. keyboard\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "2e74812d00_14", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphone case in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphone case is on the table. It is also next to the keyboard, about 11 cm away. It is further near the book, about 18 cm away." - }, - { - "id": "2e74812d00_15", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphone case in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphone case\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 11.4}, {\"object\": \"book\", \"relation\": \"near\", \"distance_cm\": 18.4}]}" - }, - { - "id": "2e74812d00_16", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the power plug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "2e74812d00_17", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "17", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. book\nB. headphone\nC. jug\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. jug" - }, - { - "id": "2e74812d00_18", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "18", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also under the jug, about 4 cm away. It is further near the book, about 39 cm away." - }, - { - "id": "2e74812d00_19", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "19", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"jug\", \"relation\": \"under\", \"distance_cm\": 3.8}, {\"object\": \"book\", \"relation\": \"near\", \"distance_cm\": 39.0}]}" - }, - { - "id": "2e74812d00_20", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "20", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "2e74812d00_21", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "21", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen in this scene?\nA. plant\nB. bottle\nC. plant pot\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. cup" - }, - { - "id": "2e74812d00_22", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "22", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen is on the table. It is also next to the cup, about 3 cm away. It is further under the plant pot, about 5 cm away." - }, - { - "id": "2e74812d00_23", - "dataset": "scannetpp", - "scene_name": "2e74812d00", - "sample_id": "23", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen\", \"support_surface\": \"table\", \"references\": [{\"object\": \"cup\", \"relation\": \"next_to\", \"distance_cm\": 2.6}, {\"object\": \"plant pot\", \"relation\": \"under\", \"distance_cm\": 4.6}]}" - }, - { - "id": "354e4a0a50_0", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the key appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "354e4a0a50_1", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. monitor\nB. computer tower\nC. speaker\nD. gamecontroller", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. gamecontroller" - }, - { - "id": "354e4a0a50_2", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also next to the gamecontroller, about 4 cm away. It is further next to the speaker, about 8 cm away." - }, - { - "id": "354e4a0a50_3", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"gamecontroller\", \"relation\": \"next_to\", \"distance_cm\": 4.0}, {\"object\": \"speaker\", \"relation\": \"next_to\", \"distance_cm\": 8.3}]}" - }, - { - "id": "354e4a0a50_4", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the usb hub appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "354e4a0a50_5", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. chair\nB. computer tower\nC. monitor\nD. headphones", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. headphones" - }, - { - "id": "354e4a0a50_6", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also near the headphones, about 18 cm away. It is further to the side of and below the monitor, about 20 cm away." - }, - { - "id": "354e4a0a50_7", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"headphones\", \"relation\": \"near\", \"distance_cm\": 18.3}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 20.2}]}" - }, - { - "id": "354e4a0a50_8", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "354e4a0a50_9", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the water bottle in this scene?\nA. fan\nB. shoe\nC. clothes\nD. plushie", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. plushie" - }, - { - "id": "354e4a0a50_10", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water bottle is to the side of and below the plushie, about 6 cm away. It is also to the side of and below the fan, about 46 cm away. It is further to the side of and below the pillow, about 46 cm away." - }, - { - "id": "354e4a0a50_11", - "dataset": "scannetpp", - "scene_name": "354e4a0a50", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water bottle\", \"support_surface\": \"bed\", \"references\": [{\"object\": \"plushie\", \"relation\": \"side_below\", \"distance_cm\": 5.6}, {\"object\": \"fan\", \"relation\": \"side_below\", \"distance_cm\": 45.9}]}" - }, - { - "id": "3583c4c70f_0", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the eyeglasses appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3583c4c70f_1", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. office chair\nB. blanket\nC. reusable shopping bag\nD. mini fridge", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. office chair" - }, - { - "id": "3583c4c70f_2", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the carpet. It is also to the side of and below the office chair, about 6 cm away. It is further to the side of and below the mini fridge, about 11 cm away." - }, - { - "id": "3583c4c70f_3", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"carpet\", \"references\": [{\"object\": \"carpet\", \"relation\": \"side_above\", \"distance_cm\": 2.4}, {\"object\": \"office chair\", \"relation\": \"side_below\", \"distance_cm\": 6.0}]}" - }, - { - "id": "3583c4c70f_4", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power bank appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3583c4c70f_5", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the extension cord in this scene?\nA. paper\nB. portable light\nC. calendar\nD. curtain", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. portable light" - }, - { - "id": "3583c4c70f_6", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the extension cord in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The extension cord is to the side of and above the portable light, about 3 cm away. It is also to the side of and above the curtain, about 6 cm away. It is further to the side of and above the calendar, about 16 cm away." - }, - { - "id": "3583c4c70f_7", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the extension cord in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"extension cord\", \"support_surface\": \"window\", \"references\": [{\"object\": \"portable light\", \"relation\": \"side_above\", \"distance_cm\": 2.8}, {\"object\": \"curtain\", \"relation\": \"side_above\", \"distance_cm\": 6.2}]}" - }, - { - "id": "3583c4c70f_8", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3583c4c70f_9", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the laptop charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The laptop charger is on the desk. It is also to the side of and above the trolley, about 3 cm away. It is further next to the books, about 3 cm away." - }, - { - "id": "3583c4c70f_10", - "dataset": "scannetpp", - "scene_name": "3583c4c70f", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the laptop charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"laptop charger\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"trolley\", \"relation\": \"side_above\", \"distance_cm\": 3.2}, {\"object\": \"books\", \"relation\": \"next_to\", \"distance_cm\": 3.3}]}" - }, - { - "id": "3597f00367_0", - "dataset": "scannetpp", - "scene_name": "3597f00367", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3597f00367_1", - "dataset": "scannetpp", - "scene_name": "3597f00367", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the extension cord in this scene?\nA. shelf\nB. skirting\nC. heater\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. table" - }, - { - "id": "3597f00367_2", - "dataset": "scannetpp", - "scene_name": "3597f00367", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the extension cord in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The extension cord is to the side of and below the table, about 9 cm away. It is also to the side of and below the shelf, about 19 cm away. It is further to the side of and below the heater, about 22 cm away." - }, - { - "id": "3597f00367_3", - "dataset": "scannetpp", - "scene_name": "3597f00367", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the extension cord in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"extension cord\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 8.8}, {\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 19.0}]}" - }, - { - "id": "35f2120068_0", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse pad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "35f2120068_1", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the extension cord in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The extension cord is to the side of and below the heater, about 9 cm away. It is also to the side of and below the heat pipe, about 9 cm away. It is further to the side of and below the document shredder, about 15 cm away." - }, - { - "id": "35f2120068_2", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the extension cord in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"extension cord\", \"support_surface\": \"heater\", \"references\": [{\"object\": \"heater\", \"relation\": \"side_below\", \"distance_cm\": 9.2}, {\"object\": \"heat pipe\", \"relation\": \"side_below\", \"distance_cm\": 9.3}]}" - }, - { - "id": "35f2120068_3", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "3", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "35f2120068_4", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "4", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the handbag in this scene?\nA. cable raceway\nB. power sockets\nC. window frame\nD. wire spool", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. power sockets" - }, - { - "id": "35f2120068_5", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the handbag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The handbag is to the side of and above the power sockets, about 13 cm away. It is also to the side of and below the window frame, about 30 cm away. It is further near the wire spool, about 34 cm away." - }, - { - "id": "35f2120068_6", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the handbag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"handbag\", \"support_surface\": \"window frame\", \"references\": [{\"object\": \"power sockets\", \"relation\": \"side_above\", \"distance_cm\": 13.3}, {\"object\": \"window frame\", \"relation\": \"side_below\", \"distance_cm\": 29.7}]}" - }, - { - "id": "35f2120068_7", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the power plug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "35f2120068_8", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "8", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also next to the electronic breadboard, about 6 cm away. It is further to the side of and below the helping hands, about 6 cm away." - }, - { - "id": "35f2120068_9", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "9", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"electronic breadboard\", \"relation\": \"next_to\", \"distance_cm\": 5.6}, {\"object\": \"helping hands\", \"relation\": \"side_below\", \"distance_cm\": 6.0}]}" - }, - { - "id": "35f2120068_10", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "10", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "35f2120068_11", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "11", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. paper\nB. box\nC. 3d printer\nD. tissue box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. paper" - }, - { - "id": "35f2120068_12", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "12", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also to the side of and above the paper, about 4 cm away. It is further to the side of and below the 3d printer, about 14 cm away." - }, - { - "id": "35f2120068_13", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "13", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"paper\", \"relation\": \"side_above\", \"distance_cm\": 4.4}, {\"object\": \"3d printer\", \"relation\": \"side_below\", \"distance_cm\": 13.5}]}" - }, - { - "id": "35f2120068_14", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "14", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue box appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "35f2120068_15", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "15", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. mug\nB. box\nC. paper\nD. thermos", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. box" - }, - { - "id": "35f2120068_16", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "16", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is on the table. It is also to the side of and above the paper, about 15 cm away. It is further near the mug, about 17 cm away." - }, - { - "id": "35f2120068_17", - "dataset": "scannetpp", - "scene_name": "35f2120068", - "sample_id": "17", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"table\", \"references\": [{\"object\": \"paper\", \"relation\": \"side_above\", \"distance_cm\": 15.0}, {\"object\": \"mug\", \"relation\": \"near\", \"distance_cm\": 16.8}]}" - }, - { - "id": "3799bd47b3_0", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the extension cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3799bd47b3_1", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. doorframe\nB. blanket\nC. bag\nD. luggage", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. luggage" - }, - { - "id": "3799bd47b3_2", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the counter. It is also to the side of and above the luggage, about 18 cm away. It is further to the side of and above the bag, about 37 cm away." - }, - { - "id": "3799bd47b3_3", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"counter\", \"references\": [{\"object\": \"luggage\", \"relation\": \"side_above\", \"distance_cm\": 18.4}, {\"object\": \"bag\", \"relation\": \"side_above\", \"distance_cm\": 36.8}]}" - }, - { - "id": "3799bd47b3_4", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the comb appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3799bd47b3_5", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. guitar\nB. curtain\nC. heater\nD. windowframe", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. guitar" - }, - { - "id": "3799bd47b3_6", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is to the side of and below the guitar, about 11 cm away. It is also to the side of and below the window, about 18 cm away. It is further to the side of and below the windowframe, about 19 cm away." - }, - { - "id": "3799bd47b3_7", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"window\", \"references\": [{\"object\": \"guitar\", \"relation\": \"side_below\", \"distance_cm\": 10.7}, {\"object\": \"window\", \"relation\": \"side_below\", \"distance_cm\": 18.0}]}" - }, - { - "id": "3799bd47b3_8", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3799bd47b3_9", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. lamp\nB. pedestal fan\nC. bookshelf\nD. power outlet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. lamp" - }, - { - "id": "3799bd47b3_10", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the lamp, about 7 cm away. It is also to the side of and below the pedestal fan, about 12 cm away. It is further to the side of and below the bookshelf, about 38 cm away." - }, - { - "id": "3799bd47b3_11", - "dataset": "scannetpp", - "scene_name": "3799bd47b3", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"lamp\", \"relation\": \"side_below\", \"distance_cm\": 7.0}, {\"object\": \"pedestal fan\", \"relation\": \"side_below\", \"distance_cm\": 12.0}]}" - }, - { - "id": "39705fbdf1_0", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the extension cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "39705fbdf1_1", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the extension cord in this scene?\nA. chair\nB. table\nC. projector\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. table" - }, - { - "id": "39705fbdf1_2", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the extension cord in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The extension cord is on the table. It is also to the side of and above the chair, about 44 cm away. It is further near the projector, about 44 cm away." - }, - { - "id": "39705fbdf1_3", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the extension cord in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"extension cord\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 43.6}, {\"object\": \"projector\", \"relation\": \"near\", \"distance_cm\": 43.9}]}" - }, - { - "id": "39705fbdf1_4", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "39705fbdf1_5", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. telephone\nB. mouse\nC. trash can\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. book" - }, - { - "id": "39705fbdf1_6", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the cabinet. It is also next to the book, about 3 cm away. It is further near the mouse, about 20 cm away." - }, - { - "id": "39705fbdf1_7", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 3.3}, {\"object\": \"mouse\", \"relation\": \"near\", \"distance_cm\": 19.8}]}" - }, - { - "id": "39705fbdf1_8", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the highlighter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "39705fbdf1_9", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. book\nB. keyboard\nC. remote\nD. telephone", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. book" - }, - { - "id": "39705fbdf1_10", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the cabinet. It is also next to the book, about 6 cm away. It is further next to the telephone, about 8 cm away." - }, - { - "id": "39705fbdf1_11", - "dataset": "scannetpp", - "scene_name": "39705fbdf1", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 6.3}, {\"object\": \"telephone\", \"relation\": \"next_to\", \"distance_cm\": 7.6}]}" - }, - { - "id": "39f36da05b_0", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the earbuds appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "39f36da05b_1", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the calculator in this scene?\nA. mouse\nB. table\nC. cup\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. cup" - }, - { - "id": "39f36da05b_2", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the calculator in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The calculator is on the table. It is also next to the cup, about 3 cm away. It is further near the mouse, about 44 cm away." - }, - { - "id": "39f36da05b_3", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the calculator in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"calculator\", \"support_surface\": \"table\", \"references\": [{\"object\": \"cup\", \"relation\": \"next_to\", \"distance_cm\": 3.3}, {\"object\": \"mouse\", \"relation\": \"near\", \"distance_cm\": 44.1}]}" - }, - { - "id": "39f36da05b_4", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the eraser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "39f36da05b_5", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is next to the cable duct, about 7 cm away. It is also next to the power socket, about 8 cm away. It is further to the side of and below the window sill, about 12 cm away." - }, - { - "id": "39f36da05b_6", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"table\", \"references\": [{\"object\": \"cable duct\", \"relation\": \"next_to\", \"distance_cm\": 6.5}, {\"object\": \"power socket\", \"relation\": \"next_to\", \"distance_cm\": 7.5}]}" - }, - { - "id": "39f36da05b_7", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "39f36da05b_8", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. cup\nB. whiteboard marker\nC. monitor\nD. whiteboard eraser", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. whiteboard marker" - }, - { - "id": "39f36da05b_9", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is next to the whiteboard marker, about 4 cm away. It is also next to the cup, about 6 cm away. It is further to the side of and below the monitor, about 6 cm away." - }, - { - "id": "39f36da05b_10", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"whiteboard marker\", \"relation\": \"next_to\", \"distance_cm\": 4.4}, {\"object\": \"cup\", \"relation\": \"next_to\", \"distance_cm\": 6.0}]}" - }, - { - "id": "39f36da05b_11", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "39f36da05b_12", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen in this scene?\nA. table\nB. headphones\nC. keyboard\nD. mouse", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "39f36da05b_13", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen is near the headphones, about 24 cm away. It is also near the mouse, about 25 cm away. It is further near the keyboard, about 32 cm away." - }, - { - "id": "39f36da05b_14", - "dataset": "scannetpp", - "scene_name": "39f36da05b", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen\", \"support_surface\": \"table\", \"references\": [{\"object\": \"headphones\", \"relation\": \"near\", \"distance_cm\": 24.0}, {\"object\": \"mouse\", \"relation\": \"near\", \"distance_cm\": 24.6}]}" - }, - { - "id": "3a3745a437_0", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3a3745a437_1", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. wall clock\nB. audio interface\nC. bag\nD. toilet paper", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. wall clock" - }, - { - "id": "3a3745a437_2", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the bookshelf. It is also to the side of and below the wall clock, about 33 cm away. It is further to the side of and above the toilet paper, about 67 cm away." - }, - { - "id": "3a3745a437_3", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"wall clock\", \"relation\": \"side_below\", \"distance_cm\": 33.0}, {\"object\": \"toilet paper\", \"relation\": \"side_above\", \"distance_cm\": 66.9}]}" - }, - { - "id": "3a3745a437_4", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3a3745a437_5", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the desk. It is also to the side of and below the shelf, about 12 cm away. It is further to the side of and below the jacket, about 12 cm away." - }, - { - "id": "3a3745a437_6", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 11.8}, {\"object\": \"jacket\", \"relation\": \"side_below\", \"distance_cm\": 11.9}]}" - }, - { - "id": "3a3745a437_7", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the notebook appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3a3745a437_8", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. bag\nB. mug\nC. shelf\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. mug" - }, - { - "id": "3a3745a437_9", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also next to the mug, about 7 cm away. It is further above the chair, about 9 cm away." - }, - { - "id": "3a3745a437_10", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"mug\", \"relation\": \"next_to\", \"distance_cm\": 6.9}, {\"object\": \"chair\", \"relation\": \"above\", \"distance_cm\": 9.0}]}" - }, - { - "id": "3a3745a437_11", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3a3745a437_12", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone charger in this scene?\nA. bed\nB. pillow\nC. mattress\nD. blanket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. mattress" - }, - { - "id": "3a3745a437_13", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is next to the mattress, about 3 cm away. It is also next to the blanket, about 10 cm away. It is further to the side of and below the pillow, about 35 cm away." - }, - { - "id": "3a3745a437_14", - "dataset": "scannetpp", - "scene_name": "3a3745a437", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"bed\", \"references\": [{\"object\": \"mattress\", \"relation\": \"next_to\", \"distance_cm\": 2.5}, {\"object\": \"blanket\", \"relation\": \"next_to\", \"distance_cm\": 9.7}]}" - }, - { - "id": "3aa115e55e_0", - "dataset": "scannetpp", - "scene_name": "3aa115e55e", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the notebook appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3aa115e55e_1", - "dataset": "scannetpp", - "scene_name": "3aa115e55e", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the remote control in this scene?\nA. cabinet\nB. power sockets\nC. speaker\nD. extension cord", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. cabinet" - }, - { - "id": "3aa115e55e_2", - "dataset": "scannetpp", - "scene_name": "3aa115e55e", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the remote control in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The remote control is on the table. It is also to the side of and above the cabinet, about 20 cm away. It is further to the side of and below the speaker, about 39 cm away." - }, - { - "id": "3aa115e55e_3", - "dataset": "scannetpp", - "scene_name": "3aa115e55e", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the remote control in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"remote control\", \"support_surface\": \"table\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_above\", \"distance_cm\": 19.8}, {\"object\": \"speaker\", \"relation\": \"side_below\", \"distance_cm\": 39.4}]}" - }, - { - "id": "3db0a1c8f3_0", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3db0a1c8f3_1", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the light switch, about 14 cm away. It is also near the beanbag, about 20 cm away. It is further to the side of and below the books, about 60 cm away." - }, - { - "id": "3db0a1c8f3_2", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"books\", \"references\": [{\"object\": \"light switch\", \"relation\": \"side_below\", \"distance_cm\": 13.9}, {\"object\": \"beanbag\", \"relation\": \"near\", \"distance_cm\": 20.5}]}" - }, - { - "id": "3db0a1c8f3_3", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "3", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3db0a1c8f3_4", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "4", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. pillar\nB. chair\nC. beanbag\nD. backpack", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. pillar" - }, - { - "id": "3db0a1c8f3_5", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also to the side of and below the pillar, about 42 cm away. It is further to the side of and above the chair, about 48 cm away." - }, - { - "id": "3db0a1c8f3_6", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"pillar\", \"relation\": \"side_below\", \"distance_cm\": 42.0}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 48.2}]}" - }, - { - "id": "3db0a1c8f3_7", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the lotion bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3db0a1c8f3_8", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headset in this scene?\nA. book\nB. piano chair\nC. backpack\nD. beanbag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. piano chair" - }, - { - "id": "3db0a1c8f3_9", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headset in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headset is to the side of and above the piano chair, about 36 cm away." - }, - { - "id": "3db0a1c8f3_10", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headset in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headset\", \"support_surface\": \"piano chair\", \"references\": [{\"object\": \"piano chair\", \"relation\": \"side_above\", \"distance_cm\": 36.3}]}" - }, - { - "id": "3db0a1c8f3_11", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3db0a1c8f3_12", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. tv receiver\nB. light switch\nC. power socket\nD. shelf", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. tv receiver" - }, - { - "id": "3db0a1c8f3_13", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is near the tv receiver, about 18 cm away. It is also to the side of and below the light switch, about 33 cm away. It is further to the side of and below the power socket, about 39 cm away." - }, - { - "id": "3db0a1c8f3_14", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"tv stand\", \"references\": [{\"object\": \"tv receiver\", \"relation\": \"near\", \"distance_cm\": 18.3}, {\"object\": \"light switch\", \"relation\": \"side_below\", \"distance_cm\": 33.4}]}" - }, - { - "id": "3db0a1c8f3_15", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3db0a1c8f3_16", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. power socket\nB. table\nC. carton\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. book" - }, - { - "id": "3db0a1c8f3_17", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is next to the book, about 7 cm away. It is also near the power socket, about 27 cm away. It is further to the side of and below the table, about 29 cm away." - }, - { - "id": "3db0a1c8f3_18", - "dataset": "scannetpp", - "scene_name": "3db0a1c8f3", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 6.6}, {\"object\": \"power socket\", \"relation\": \"near\", \"distance_cm\": 26.8}]}" - }, - { - "id": "3eba679830_0", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3eba679830_1", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. cabinet\nB. file folder\nC. office chair\nD. tissue box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. cabinet" - }, - { - "id": "3eba679830_2", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the cabinet, about 6 cm away. It is also to the side of and below the tissue box, about 18 cm away. It is further to the side of and below the office chair, about 19 cm away." - }, - { - "id": "3eba679830_3", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 6.3}, {\"object\": \"tissue box\", \"relation\": \"side_below\", \"distance_cm\": 18.3}]}" - }, - { - "id": "3eba679830_4", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3eba679830_5", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. mouse\nB. monitor\nC. office chair\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. mouse" - }, - { - "id": "3eba679830_6", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also near the mouse, about 16 cm away. It is further to the side of and below the monitor, about 29 cm away." - }, - { - "id": "3eba679830_7", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"mouse\", \"relation\": \"near\", \"distance_cm\": 15.6}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 29.1}]}" - }, - { - "id": "3eba679830_8", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the tote bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3eba679830_9", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. lamp\nB. curtain\nC. monitor\nD. keyboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. monitor" - }, - { - "id": "3eba679830_10", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also to the side of and below the monitor, about 6 cm away. It is further near the keyboard, about 16 cm away." - }, - { - "id": "3eba679830_11", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 6.5}, {\"object\": \"keyboard\", \"relation\": \"near\", \"distance_cm\": 15.6}]}" - }, - { - "id": "3eba679830_12", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3eba679830_13", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. backpack\nB. power outlet\nC. café curtains\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. power outlet" - }, - { - "id": "3eba679830_14", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the power outlet, about 48 cm away." - }, - { - "id": "3eba679830_15", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"vacuum cleaner\", \"references\": [{\"object\": \"power outlet\", \"relation\": \"side_below\", \"distance_cm\": 48.0}]}" - }, - { - "id": "3eba679830_16", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the comb appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3eba679830_17", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "17", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. office chair\nB. file folder\nC. backpack\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. file folder" - }, - { - "id": "3eba679830_18", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "18", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is on the cabinet. It is also next to the file folder, about 6 cm away. It is further to the side of and above the backpack, about 18 cm away." - }, - { - "id": "3eba679830_19", - "dataset": "scannetpp", - "scene_name": "3eba679830", - "sample_id": "19", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"file folder\", \"relation\": \"next_to\", \"distance_cm\": 6.5}, {\"object\": \"backpack\", \"relation\": \"side_above\", \"distance_cm\": 18.3}]}" - }, - { - "id": "3ee9d33aef_0", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cable appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3ee9d33aef_1", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cable in this scene?\nA. nightstand\nB. clothes\nC. power strip\nD. power adapter", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. power strip" - }, - { - "id": "3ee9d33aef_2", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cable in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cable is next to the power strip, about 10 cm away. It is also to the side of and below the power adapter, about 14 cm away. It is further to the side of and below the nightstand, about 21 cm away." - }, - { - "id": "3ee9d33aef_3", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cable in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cable\", \"support_surface\": \"nightstand\", \"references\": [{\"object\": \"power strip\", \"relation\": \"next_to\", \"distance_cm\": 10.0}, {\"object\": \"power adapter\", \"relation\": \"side_below\", \"distance_cm\": 13.7}]}" - }, - { - "id": "3ee9d33aef_4", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the earbuds appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3ee9d33aef_5", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. pen stand\nB. split\nC. bottle\nD. tissue box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. split" - }, - { - "id": "3ee9d33aef_6", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the nightstand. It is also next to the split. It is further under the bottle, about 5 cm away." - }, - { - "id": "3ee9d33aef_7", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"nightstand\", \"references\": [{\"object\": \"split\", \"relation\": \"next_to\", \"distance_cm\": 2.0}, {\"object\": \"bottle\", \"relation\": \"under\", \"distance_cm\": 5.0}]}" - }, - { - "id": "3ee9d33aef_8", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3ee9d33aef_9", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. cable\nB. office chair\nC. office table\nD. nightstand", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. cable" - }, - { - "id": "3ee9d33aef_10", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is next to the cable, about 10 cm away. It is also to the side of and below the office table, about 12 cm away. It is further to the side of and below the office chair, about 26 cm away." - }, - { - "id": "3ee9d33aef_11", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"office table\", \"references\": [{\"object\": \"cable\", \"relation\": \"next_to\", \"distance_cm\": 10.0}, {\"object\": \"office table\", \"relation\": \"side_below\", \"distance_cm\": 11.7}]}" - }, - { - "id": "3ee9d33aef_12", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the marker appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3ee9d33aef_13", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. pen stand\nB. book\nC. split\nD. headphones", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. pen stand" - }, - { - "id": "3ee9d33aef_14", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is on the nightstand. It is also under the pen stand. It is further to the side of and above the book, about 4 cm away." - }, - { - "id": "3ee9d33aef_15", - "dataset": "scannetpp", - "scene_name": "3ee9d33aef", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"nightstand\", \"references\": [{\"object\": \"pen stand\", \"relation\": \"under\", \"distance_cm\": 2.1}, {\"object\": \"book\", \"relation\": \"side_above\", \"distance_cm\": 3.5}]}" - }, - { - "id": "3f15a9266d_0", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the earphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3f15a9266d_1", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. mouse\nB. book\nC. folder\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. mouse" - }, - { - "id": "3f15a9266d_2", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also next to the mouse, about 8 cm away. It is further next to the book, about 10 cm away." - }, - { - "id": "3f15a9266d_3", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 8.3}, {\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 10.4}]}" - }, - { - "id": "3f15a9266d_4", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3f15a9266d_5", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. chair\nB. keyboard\nC. monitor\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. keyboard" - }, - { - "id": "3f15a9266d_6", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is next to the keyboard, about 8 cm away. It is also to the side of and below the monitor, about 28 cm away. It is further to the side of and above the chair, about 30 cm away." - }, - { - "id": "3f15a9266d_7", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 8.3}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 27.8}]}" - }, - { - "id": "3f15a9266d_8", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the sticky note appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3f15a9266d_9", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse pad in this scene?\nA. book\nB. folder\nC. chair\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. chair" - }, - { - "id": "3f15a9266d_10", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse pad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse pad is to the side of and above the chair, about 24 cm away. It is also to the side of and below the monitor, about 26 cm away. It is further near the book, about 50 cm away." - }, - { - "id": "3f15a9266d_11", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse pad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse pad\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 24.2}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 25.7}]}" - }, - { - "id": "3f15a9266d_12", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3f15a9266d_13", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pencil holder in this scene?\nA. cup\nB. book\nC. telephone\nD. picture", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. cup" - }, - { - "id": "3f15a9266d_14", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pencil holder in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pencil holder is next to the cup, about 3 cm away. It is also to the side of and below the window, about 7 cm away. It is further to the side of and below the picture, about 10 cm away." - }, - { - "id": "3f15a9266d_15", - "dataset": "scannetpp", - "scene_name": "3f15a9266d", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pencil holder in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pencil holder\", \"support_surface\": \"window\", \"references\": [{\"object\": \"cup\", \"relation\": \"next_to\", \"distance_cm\": 3.4}, {\"object\": \"window\", \"relation\": \"side_below\", \"distance_cm\": 7.4}]}" - }, - { - "id": "3f1e1610de_0", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the marker appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3f1e1610de_1", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. exercise ball\nB. speaker\nC. monitor\nD. mouse", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. exercise ball" - }, - { - "id": "3f1e1610de_2", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also to the side of and above the exercise ball, about 7 cm away. It is further next to the mouse, about 10 cm away." - }, - { - "id": "3f1e1610de_3", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"exercise ball\", \"relation\": \"side_above\", \"distance_cm\": 6.6}, {\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 9.6}]}" - }, - { - "id": "3f1e1610de_4", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "3f1e1610de_5", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. cabinet\nB. exercise ball\nC. desk\nD. keyboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. desk" - }, - { - "id": "3f1e1610de_6", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is to the side of and above the desk, about 8 cm away. It is also next to the keyboard, about 10 cm away. It is further to the side of and above the cabinet, about 19 cm away." - }, - { - "id": "3f1e1610de_7", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"desk\", \"relation\": \"side_above\", \"distance_cm\": 7.9}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 9.6}]}" - }, - { - "id": "3f1e1610de_8", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the razor appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "3f1e1610de_9", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse pad in this scene?\nA. exercise ball\nB. speaker\nC. monitor\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. cabinet" - }, - { - "id": "3f1e1610de_10", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse pad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse pad is on the desk. It is also to the side of and above the cabinet, about 14 cm away. It is further to the side of and above the exercise ball, about 18 cm away." - }, - { - "id": "3f1e1610de_11", - "dataset": "scannetpp", - "scene_name": "3f1e1610de", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse pad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse pad\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_above\", \"distance_cm\": 14.0}, {\"object\": \"exercise ball\", \"relation\": \"side_above\", \"distance_cm\": 18.1}]}" - }, - { - "id": "40b56bf310_0", - "dataset": "scannetpp", - "scene_name": "40b56bf310", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the tote bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "40b56bf310_1", - "dataset": "scannetpp", - "scene_name": "40b56bf310", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. cabinet\nB. socket\nC. column\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. monitor" - }, - { - "id": "40b56bf310_2", - "dataset": "scannetpp", - "scene_name": "40b56bf310", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also to the side of and below the monitor, about 18 cm away. It is further near the socket, about 19 cm away." - }, - { - "id": "40b56bf310_3", - "dataset": "scannetpp", - "scene_name": "40b56bf310", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 17.8}, {\"object\": \"socket\", \"relation\": \"near\", \"distance_cm\": 19.2}]}" - }, - { - "id": "413085a827_0", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "413085a827_1", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. jar\nB. office chair\nC. plate\nD. mug", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. mug" - }, - { - "id": "413085a827_2", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also to the side of and above the office chair, about 13 cm away. It is further to the side of and above the plate, about 27 cm away." - }, - { - "id": "413085a827_3", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 13.4}, {\"object\": \"plate\", \"relation\": \"side_above\", \"distance_cm\": 26.5}]}" - }, - { - "id": "413085a827_4", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "413085a827_5", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. jackets\nB. bottle\nC. puch\nD. paper bag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. bottle" - }, - { - "id": "413085a827_6", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is under the bottle, about 3 cm away. It is also near the puch, about 24 cm away. It is further to the side of and above the jackets, about 31 cm away." - }, - { - "id": "413085a827_7", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"bag\", \"references\": [{\"object\": \"bottle\", \"relation\": \"under\", \"distance_cm\": 3.0}, {\"object\": \"puch\", \"relation\": \"near\", \"distance_cm\": 24.4}]}" - }, - { - "id": "413085a827_8", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the smartphone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "413085a827_9", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power extension in this scene?\nA. heater\nB. table\nC. power socket\nD. electric socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. table" - }, - { - "id": "413085a827_10", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power extension in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power extension is to the side of and below the table, about 21 cm away. It is also to the side of and below the power socket, about 32 cm away. It is further to the side of and below the heater, about 52 cm away." - }, - { - "id": "413085a827_11", - "dataset": "scannetpp", - "scene_name": "413085a827", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power extension in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power extension\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 20.8}, {\"object\": \"power socket\", \"relation\": \"side_below\", \"distance_cm\": 31.7}]}" - }, - { - "id": "43cd995c51_0", - "dataset": "scannetpp", - "scene_name": "43cd995c51", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the adapter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "43cd995c51_1", - "dataset": "scannetpp", - "scene_name": "43cd995c51", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. tray\nB. books\nC. table\nD. desk lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. table" - }, - { - "id": "43cd995c51_2", - "dataset": "scannetpp", - "scene_name": "43cd995c51", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is to the side of and above the table, about 22 cm away. It is also to the side of and above the tray, about 44 cm away. It is further to the side of and above the books, about 53 cm away." - }, - { - "id": "43cd995c51_3", - "dataset": "scannetpp", - "scene_name": "43cd995c51", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 22.4}, {\"object\": \"tray\", \"relation\": \"side_above\", \"distance_cm\": 43.8}]}" - }, - { - "id": "4422722c49_0", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "4422722c49_1", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the charger in this scene?\nA. power strip\nB. box\nC. wifi router\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. box" - }, - { - "id": "4422722c49_2", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The charger is on the table. It is also next to the box, about 7 cm away. It is further next to the power strip, about 8 cm away." - }, - { - "id": "4422722c49_3", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"charger\", \"support_surface\": \"table\", \"references\": [{\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 7.1}, {\"object\": \"power strip\", \"relation\": \"next_to\", \"distance_cm\": 8.2}]}" - }, - { - "id": "4422722c49_4", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the glue bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "4422722c49_5", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. box\nB. shelf\nC. book\nD. file binder", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. book" - }, - { - "id": "4422722c49_6", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is to the side of and below the file binder, about 10 cm away. It is also to the side of and below the box, about 40 cm away. It is further near the paper towel dispenser, about 50 cm away." - }, - { - "id": "4422722c49_7", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"file binder\", \"relation\": \"side_below\", \"distance_cm\": 9.8}, {\"object\": \"box\", \"relation\": \"side_below\", \"distance_cm\": 39.7}]}" - }, - { - "id": "4422722c49_8", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "4422722c49_9", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also next to the keyboard, about 11 cm away. It is further next to the box, about 11 cm away." - }, - { - "id": "4422722c49_10", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 10.5}, {\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 10.8}]}" - }, - { - "id": "4422722c49_11", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "4422722c49_12", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. computer tower\nB. folder\nC. keyboard\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. computer tower" - }, - { - "id": "4422722c49_13", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is on the table. It is also to the side of and below the computer tower, about 9 cm away. It is further to the side of and below the monitor, about 25 cm away." - }, - { - "id": "4422722c49_14", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"computer tower\", \"relation\": \"side_below\", \"distance_cm\": 9.3}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 25.1}]}" - }, - { - "id": "4422722c49_15", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue box appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "4422722c49_16", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. laboratory pellet press\nB. office chair\nC. step stool\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. table" - }, - { - "id": "4422722c49_17", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is on the table. It is also to the side of and above the office chair, about 18 cm away. It is further to the side of and below the laboratory pellet press, about 28 cm away." - }, - { - "id": "4422722c49_18", - "dataset": "scannetpp", - "scene_name": "4422722c49", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"table\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 17.6}, {\"object\": \"laboratory pellet press\", \"relation\": \"side_below\", \"distance_cm\": 27.9}]}" - }, - { - "id": "4423a61d09_0", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "4423a61d09_1", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. table\nB. monitor\nC. telephone\nD. mouse pad", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "4423a61d09_2", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is to the side of and above the table. It is also next to the telephone, about 3 cm away. It is further next to the mouse pad, about 13 cm away." - }, - { - "id": "4423a61d09_3", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 2.4}, {\"object\": \"telephone\", \"relation\": \"next_to\", \"distance_cm\": 3.5}]}" - }, - { - "id": "4423a61d09_4", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphone case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "4423a61d09_5", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse pad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse pad is on the table. It is also next to the file binder, about 6 cm away. It is further to the side of and below the monitor, about 6 cm away." - }, - { - "id": "4423a61d09_6", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse pad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse pad\", \"support_surface\": \"table\", \"references\": [{\"object\": \"file binder\", \"relation\": \"next_to\", \"distance_cm\": 6.0}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 6.4}]}" - }, - { - "id": "4423a61d09_7", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the power extension appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "4423a61d09_8", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. monitor\nB. laptop\nC. cardboard box\nD. table fan", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. monitor" - }, - { - "id": "4423a61d09_9", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the monitor, about 16 cm away. It is also to the side of and below the table fan, about 22 cm away. It is further to the side of and above the cardboard box, about 24 cm away." - }, - { - "id": "4423a61d09_10", - "dataset": "scannetpp", - "scene_name": "4423a61d09", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 15.6}, {\"object\": \"table fan\", \"relation\": \"side_below\", \"distance_cm\": 21.6}]}" - }, - { - "id": "442b144761_0", - "dataset": "scannetpp", - "scene_name": "442b144761", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the smartphone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "442b144761_1", - "dataset": "scannetpp", - "scene_name": "442b144761", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. markers\nB. chair\nC. power sockets\nD. whiteboard eraser", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. power sockets" - }, - { - "id": "442b144761_2", - "dataset": "scannetpp", - "scene_name": "442b144761", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is next to the power sockets, about 5 cm away. It is also to the side of and above the markers, about 33 cm away. It is further to the side of and above the whiteboard eraser, about 50 cm away." - }, - { - "id": "442b144761_3", - "dataset": "scannetpp", - "scene_name": "442b144761", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"conference table\", \"references\": [{\"object\": \"power sockets\", \"relation\": \"next_to\", \"distance_cm\": 5.2}, {\"object\": \"markers\", \"relation\": \"side_above\", \"distance_cm\": 33.2}]}" - }, - { - "id": "469f112e38_0", - "dataset": "scannetpp", - "scene_name": "469f112e38", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the spray bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "469f112e38_1", - "dataset": "scannetpp", - "scene_name": "469f112e38", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is next to the frying pan, about 13 cm away. It is also next to the blender, about 13 cm away. It is further near the bowl, about 15 cm away." - }, - { - "id": "469f112e38_2", - "dataset": "scannetpp", - "scene_name": "469f112e38", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"window\", \"references\": [{\"object\": \"frying pan\", \"relation\": \"next_to\", \"distance_cm\": 12.7}, {\"object\": \"blender\", \"relation\": \"next_to\", \"distance_cm\": 12.8}]}" - }, - { - "id": "469f112e38_3", - "dataset": "scannetpp", - "scene_name": "469f112e38", - "sample_id": "3", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "469f112e38_4", - "dataset": "scannetpp", - "scene_name": "469f112e38", - "sample_id": "4", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tote bag in this scene?\nA. blender\nB. chair\nC. bottle\nD. bowl", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. chair" - }, - { - "id": "469f112e38_5", - "dataset": "scannetpp", - "scene_name": "469f112e38", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tote bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tote bag is to the side of and below the chair, about 65 cm away." - }, - { - "id": "469f112e38_6", - "dataset": "scannetpp", - "scene_name": "469f112e38", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tote bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tote bag\", \"support_surface\": \"office chair\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 65.3}]}" - }, - { - "id": "47c1209b9e_0", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the hairbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "47c1209b9e_1", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. bowl\nB. books\nC. router\nD. shelf", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. shelf" - }, - { - "id": "47c1209b9e_2", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the shelf, about 15 cm away. It is also to the side of and below the books, about 25 cm away. It is further to the side of and below the bowl, about 46 cm away." - }, - { - "id": "47c1209b9e_3", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 15.2}, {\"object\": \"books\", \"relation\": \"side_below\", \"distance_cm\": 24.7}]}" - }, - { - "id": "47c1209b9e_4", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "47c1209b9e_5", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the laptop charger in this scene?\nA. mouse\nB. book\nC. mouse pad\nD. table lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. mouse pad" - }, - { - "id": "47c1209b9e_6", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the laptop charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The laptop charger is on the desk. It is also next to the mouse pad, about 4 cm away. It is further next to the book, about 9 cm away." - }, - { - "id": "47c1209b9e_7", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the laptop charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"laptop charger\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"mouse pad\", \"relation\": \"next_to\", \"distance_cm\": 4.1}, {\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 8.8}]}" - }, - { - "id": "47c1209b9e_8", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the lanyard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "47c1209b9e_9", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also next to the laptop charger, about 9 cm away. It is further to the side of and below the monitor, about 10 cm away." - }, - { - "id": "47c1209b9e_10", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"laptop charger\", \"relation\": \"next_to\", \"distance_cm\": 9.3}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 10.0}]}" - }, - { - "id": "47c1209b9e_11", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse pad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "47c1209b9e_12", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "12", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse pad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse pad is on the desk. It is also next to the laptop charger, about 4 cm away. It is further under the monitor, about 5 cm away." - }, - { - "id": "47c1209b9e_13", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "13", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse pad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse pad\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"laptop charger\", \"relation\": \"next_to\", \"distance_cm\": 4.1}, {\"object\": \"monitor\", \"relation\": \"under\", \"distance_cm\": 5.0}]}" - }, - { - "id": "47c1209b9e_14", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "14", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "47c1209b9e_15", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "15", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. power socket\nB. shelf rail\nC. jar\nD. mayonnaise bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. shelf rail" - }, - { - "id": "47c1209b9e_16", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "16", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is under the shelf rail. It is also to the side of and above the power socket, about 9 cm away. It is further to the side of and below the mayonnaise bottle, about 13 cm away." - }, - { - "id": "47c1209b9e_17", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "17", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"bedside table\", \"references\": [{\"object\": \"shelf rail\", \"relation\": \"under\", \"distance_cm\": 2.4}, {\"object\": \"power socket\", \"relation\": \"side_above\", \"distance_cm\": 8.9}]}" - }, - { - "id": "47c1209b9e_18", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "18", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "47c1209b9e_19", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "19", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the water bottle in this scene?\nA. bedside table\nB. mayonnaise bottle\nC. microwave\nD. jar", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. microwave" - }, - { - "id": "47c1209b9e_20", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "20", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water bottle is next to the microwave, about 5 cm away. It is also to the side of and above the jar, about 35 cm away. It is further to the side of and above the mayonnaise bottle, about 36 cm away." - }, - { - "id": "47c1209b9e_21", - "dataset": "scannetpp", - "scene_name": "47c1209b9e", - "sample_id": "21", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water bottle\", \"support_surface\": \"bedside table\", \"references\": [{\"object\": \"microwave\", \"relation\": \"next_to\", \"distance_cm\": 4.5}, {\"object\": \"jar\", \"relation\": \"side_above\", \"distance_cm\": 35.3}]}" - }, - { - "id": "48573f4c95_0", - "dataset": "scannetpp", - "scene_name": "48573f4c95", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "48573f4c95_1", - "dataset": "scannetpp", - "scene_name": "48573f4c95", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. office table\nB. stapler\nC. pens\nD. paper", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. pens" - }, - { - "id": "48573f4c95_2", - "dataset": "scannetpp", - "scene_name": "48573f4c95", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is to the side of and above the office table, about 6 cm away. It is also next to the stapler, about 8 cm away. It is further to the side of and above the paper, about 14 cm away." - }, - { - "id": "48573f4c95_3", - "dataset": "scannetpp", - "scene_name": "48573f4c95", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"pen stand\", \"references\": [{\"object\": \"office table\", \"relation\": \"side_above\", \"distance_cm\": 5.6}, {\"object\": \"stapler\", \"relation\": \"next_to\", \"distance_cm\": 8.4}]}" - }, - { - "id": "48701abb21_0", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "48701abb21_1", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. cable\nB. cooking pan\nC. cooking pot\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. cabinet" - }, - { - "id": "48701abb21_2", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is near the cabinet, about 16 cm away. It is also near the cooking pot, about 18 cm away. It is further to the side of and above the cooking pan, about 30 cm away." - }, - { - "id": "48701abb21_3", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"near\", \"distance_cm\": 15.8}, {\"object\": \"cooking pot\", \"relation\": \"near\", \"distance_cm\": 17.9}]}" - }, - { - "id": "48701abb21_4", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the glasses appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "48701abb21_5", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cable in this scene?\nA. cooking pan\nB. bottle\nC. cabinet\nD. ceiling lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. ceiling lamp" - }, - { - "id": "48701abb21_6", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cable in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cable is above the ceiling lamp." - }, - { - "id": "48701abb21_7", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cable in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cable\", \"support_surface\": \"ceiling lamp\", \"references\": [{\"object\": \"ceiling lamp\", \"relation\": \"above\", \"distance_cm\": 2.4}]}" - }, - { - "id": "48701abb21_8", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the spray bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "48701abb21_9", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the spray bottle in this scene?\nA. kettle\nB. knife\nC. kitchen cabinet\nD. water jug", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. knife" - }, - { - "id": "48701abb21_10", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is to the side of and below the knife, about 5 cm away. It is also near the water jug, about 28 cm away. It is further to the side of and below the kitchen cabinet, about 39 cm away." - }, - { - "id": "48701abb21_11", - "dataset": "scannetpp", - "scene_name": "48701abb21", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"kitchen cabinet\", \"references\": [{\"object\": \"knife\", \"relation\": \"side_below\", \"distance_cm\": 5.5}, {\"object\": \"water jug\", \"relation\": \"near\", \"distance_cm\": 28.5}]}" - }, - { - "id": "48da625e8c_0", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power bank appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "48da625e8c_1", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. book\nB. mouse\nC. tablet stand\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. table" - }, - { - "id": "48da625e8c_2", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the table, about 8 cm away. It is also to the side of and below the tablet stand, about 12 cm away. It is further near the book, about 16 cm away." - }, - { - "id": "48da625e8c_3", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"desk lamp\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 7.6}, {\"object\": \"tablet stand\", \"relation\": \"side_below\", \"distance_cm\": 11.8}]}" - }, - { - "id": "48da625e8c_4", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "48da625e8c_5", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. book\nB. plant\nC. box\nD. tablet stand", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. book" - }, - { - "id": "48da625e8c_6", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also next to the box, about 7 cm away. It is further to the side of and below the plant, about 18 cm away." - }, - { - "id": "48da625e8c_7", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 6.9}, {\"object\": \"plant\", \"relation\": \"side_below\", \"distance_cm\": 18.1}]}" - }, - { - "id": "48da625e8c_8", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "48da625e8c_9", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. tablet stand\nB. desk lamp\nC. box\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. box" - }, - { - "id": "48da625e8c_10", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also next to the box, about 5 cm away. It is further to the side of and below the desk lamp, about 7 cm away." - }, - { - "id": "48da625e8c_11", - "dataset": "scannetpp", - "scene_name": "48da625e8c", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 4.5}, {\"object\": \"desk lamp\", \"relation\": \"side_below\", \"distance_cm\": 6.8}]}" - }, - { - "id": "49a82360aa_0", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "49a82360aa_1", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. bottle\nB. phone stand\nC. windowsill\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. phone stand" - }, - { - "id": "49a82360aa_2", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also next to the phone stand, about 7 cm away. It is further near the windowsill, about 21 cm away." - }, - { - "id": "49a82360aa_3", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"phone stand\", \"relation\": \"next_to\", \"distance_cm\": 6.8}, {\"object\": \"windowsill\", \"relation\": \"near\", \"distance_cm\": 21.0}]}" - }, - { - "id": "49a82360aa_4", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power bank appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "49a82360aa_5", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also next to the keyboard, about 5 cm away. It is further next to the laptop, about 6 cm away." - }, - { - "id": "49a82360aa_6", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 5.5}, {\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 5.7}]}" - }, - { - "id": "49a82360aa_7", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "49a82360aa_8", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone stand in this scene?\nA. windowsill\nB. paper towel\nC. cup\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. paper towel" - }, - { - "id": "49a82360aa_9", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone stand in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone stand is on the table. It is also next to the paper towel, about 3 cm away. It is further next to the cup, about 7 cm away." - }, - { - "id": "49a82360aa_10", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone stand in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone stand\", \"support_surface\": \"table\", \"references\": [{\"object\": \"paper towel\", \"relation\": \"next_to\", \"distance_cm\": 3.0}, {\"object\": \"cup\", \"relation\": \"next_to\", \"distance_cm\": 6.8}]}" - }, - { - "id": "49a82360aa_11", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "49a82360aa_12", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. tea pot\nB. tablet\nC. basket\nD. tv stand", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. basket" - }, - { - "id": "49a82360aa_13", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and above the basket, about 15 cm away. It is also near the tablet, about 19 cm away. It is further to the side of and below the tea pot, about 26 cm away." - }, - { - "id": "49a82360aa_14", - "dataset": "scannetpp", - "scene_name": "49a82360aa", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"basket\", \"relation\": \"side_above\", \"distance_cm\": 15.1}, {\"object\": \"tablet\", \"relation\": \"near\", \"distance_cm\": 18.9}]}" - }, - { - "id": "54b6127146_0", - "dataset": "scannetpp", - "scene_name": "54b6127146", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "54b6127146_1", - "dataset": "scannetpp", - "scene_name": "54b6127146", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. telephone\nB. case\nC. nightstand\nD. suitcase", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. suitcase" - }, - { - "id": "54b6127146_2", - "dataset": "scannetpp", - "scene_name": "54b6127146", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the suitcase, about 43 cm away. It is also to the side of and below the nightstand, about 54 cm away. It is further to the side of and below the telephone, about 63 cm away." - }, - { - "id": "54b6127146_3", - "dataset": "scannetpp", - "scene_name": "54b6127146", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"nightstand\", \"references\": [{\"object\": \"suitcase\", \"relation\": \"side_below\", \"distance_cm\": 43.1}, {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 54.2}]}" - }, - { - "id": "54b6127146_4", - "dataset": "scannetpp", - "scene_name": "54b6127146", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "54b6127146_5", - "dataset": "scannetpp", - "scene_name": "54b6127146", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. painting\nB. towel\nC. sink\nD. toilet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. towel" - }, - { - "id": "54b6127146_6", - "dataset": "scannetpp", - "scene_name": "54b6127146", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is on the counter. It is also to the side of and above the towel, about 15 cm away. It is further to the side of and above the toilet, about 29 cm away." - }, - { - "id": "54b6127146_7", - "dataset": "scannetpp", - "scene_name": "54b6127146", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"counter\", \"references\": [{\"object\": \"towel\", \"relation\": \"side_above\", \"distance_cm\": 15.3}, {\"object\": \"toilet\", \"relation\": \"side_above\", \"distance_cm\": 29.1}]}" - }, - { - "id": "5748ce6f01_0", - "dataset": "scannetpp", - "scene_name": "5748ce6f01", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the calculator appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "5748ce6f01_1", - "dataset": "scannetpp", - "scene_name": "5748ce6f01", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. office chair\nB. monitor\nC. table\nD. keyboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. office chair" - }, - { - "id": "5748ce6f01_2", - "dataset": "scannetpp", - "scene_name": "5748ce6f01", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also to the side of and above the office chair, about 50 cm away." - }, - { - "id": "5748ce6f01_3", - "dataset": "scannetpp", - "scene_name": "5748ce6f01", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 50.0}]}" - }, - { - "id": "57d9fecae9_0", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the earbuds appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "57d9fecae9_1", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cable in this scene?\nA. power strip\nB. hanging storage\nC. basket\nD. coax outlet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. power strip" - }, - { - "id": "57d9fecae9_2", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cable in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cable is next to the power strip, about 3 cm away. It is also to the side of and below the hanging storage, about 9 cm away. It is further to the side of and below the coax outlet, about 10 cm away." - }, - { - "id": "57d9fecae9_3", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cable in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cable\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"power strip\", \"relation\": \"next_to\", \"distance_cm\": 3.0}, {\"object\": \"hanging storage\", \"relation\": \"side_below\", \"distance_cm\": 9.4}]}" - }, - { - "id": "57d9fecae9_4", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "57d9fecae9_5", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the notebook in this scene?\nA. laptop\nB. monitor\nC. boxes\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. boxes" - }, - { - "id": "57d9fecae9_6", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the notebook in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The notebook is on the desk. It is also to the side of and above the chair, about 17 cm away. It is further to the side of and below the monitor, about 18 cm away." - }, - { - "id": "57d9fecae9_7", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the notebook in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"notebook\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 17.4}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 18.3}]}" - }, - { - "id": "57d9fecae9_8", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the scissors appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "57d9fecae9_9", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the scissors in this scene?\nA. doorframe\nB. sponge\nC. pan\nD. kitchen counter", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. pan" - }, - { - "id": "57d9fecae9_10", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the scissors in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The scissors is to the side of and above the pan, about 19 cm away. It is also to the side of and above the doorframe, about 25 cm away. It is further to the side of and above the door, about 25 cm away." - }, - { - "id": "57d9fecae9_11", - "dataset": "scannetpp", - "scene_name": "57d9fecae9", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the scissors in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"scissors\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"pan\", \"relation\": \"side_above\", \"distance_cm\": 18.6}, {\"object\": \"doorframe\", \"relation\": \"side_above\", \"distance_cm\": 25.0}]}" - }, - { - "id": "5f0fb991a7_0", - "dataset": "scannetpp", - "scene_name": "5f0fb991a7", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the glue bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "5f0fb991a7_1", - "dataset": "scannetpp", - "scene_name": "5f0fb991a7", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the desk, about 8 cm away. It is also to the side of and above the garment bag, about 8 cm away. It is further to the side of and above the luggage, about 16 cm away." - }, - { - "id": "5f0fb991a7_2", - "dataset": "scannetpp", - "scene_name": "5f0fb991a7", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"desk\", \"relation\": \"side_below\", \"distance_cm\": 7.9}, {\"object\": \"garment bag\", \"relation\": \"side_above\", \"distance_cm\": 8.0}]}" - }, - { - "id": "5f0fb991a7_3", - "dataset": "scannetpp", - "scene_name": "5f0fb991a7", - "sample_id": "3", - "question_type": "target presence verification", - "question": "In the entire scene, did the toothbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "5f0fb991a7_4", - "dataset": "scannetpp", - "scene_name": "5f0fb991a7", - "sample_id": "4", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toothbrush in this scene?\nA. sink countertop\nB. light switch\nC. sink\nD. towel", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. towel" - }, - { - "id": "5f0fb991a7_5", - "dataset": "scannetpp", - "scene_name": "5f0fb991a7", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toothbrush in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toothbrush is on the shelf. It is also to the side of and above the towel, about 36 cm away. It is further to the side of and above the sink countertop, about 40 cm away." - }, - { - "id": "5f0fb991a7_6", - "dataset": "scannetpp", - "scene_name": "5f0fb991a7", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toothbrush in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toothbrush\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"towel\", \"relation\": \"side_above\", \"distance_cm\": 36.2}, {\"object\": \"sink countertop\", \"relation\": \"side_above\", \"distance_cm\": 40.3}]}" - }, - { - "id": "610eff6509_0", - "dataset": "scannetpp", - "scene_name": "610eff6509", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the marker appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "610eff6509_1", - "dataset": "scannetpp", - "scene_name": "610eff6509", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. windowsill\nB. window frame\nC. chopping boards\nD. bowl", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bowl" - }, - { - "id": "610eff6509_2", - "dataset": "scannetpp", - "scene_name": "610eff6509", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the bowl, about 6 cm away. It is also next to the windowsill, about 13 cm away. It is further to the side of and below the window frame, about 17 cm away." - }, - { - "id": "610eff6509_3", - "dataset": "scannetpp", - "scene_name": "610eff6509", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"kitchen cabinet\", \"references\": [{\"object\": \"bowl\", \"relation\": \"side_above\", \"distance_cm\": 6.1}, {\"object\": \"windowsill\", \"relation\": \"next_to\", \"distance_cm\": 13.2}]}" - }, - { - "id": "612f70fe00_0", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "612f70fe00_1", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. baguettes\nB. cutting board\nC. cleaning wipes\nD. salt", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. cleaning wipes" - }, - { - "id": "612f70fe00_2", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also next to the cleaning wipes, about 5 cm away. It is further next to the baguettes, about 7 cm away." - }, - { - "id": "612f70fe00_3", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"cleaning wipes\", \"relation\": \"next_to\", \"distance_cm\": 4.8}, {\"object\": \"baguettes\", \"relation\": \"next_to\", \"distance_cm\": 6.5}]}" - }, - { - "id": "612f70fe00_4", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the smartphone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "612f70fe00_5", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the spray bottle in this scene?\nA. paper bag\nB. detergent\nC. air fryer\nD. shelf rail", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. shelf rail" - }, - { - "id": "612f70fe00_6", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is to the side of and above the shelf rail, about 28 cm away. It is also to the side of and above the paper bag, about 73 cm away. It is further to the side of and above the detergent, about 77 cm away." - }, - { - "id": "612f70fe00_7", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"shelf rail\", \"references\": [{\"object\": \"shelf rail\", \"relation\": \"side_above\", \"distance_cm\": 28.4}, {\"object\": \"paper bag\", \"relation\": \"side_above\", \"distance_cm\": 72.5}]}" - }, - { - "id": "612f70fe00_8", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the stapler appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "612f70fe00_9", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tote bag in this scene?\nA. table\nB. cutting board\nC. baguettes\nD. knife", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "612f70fe00_10", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tote bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tote bag is under the table, about 4 cm away. It is also to the side of and below the cutting board, about 6 cm away. It is further to the side of and below the knife, about 11 cm away." - }, - { - "id": "612f70fe00_11", - "dataset": "scannetpp", - "scene_name": "612f70fe00", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tote bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tote bag\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"under\", \"distance_cm\": 4.5}, {\"object\": \"cutting board\", \"relation\": \"side_below\", \"distance_cm\": 6.3}]}" - }, - { - "id": "618310ed87_0", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "618310ed87_1", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. power socket\nB. pot\nC. picture\nD. light switch", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. pot" - }, - { - "id": "618310ed87_2", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the pot, about 28 cm away. It is also to the side of and below the picture, about 33 cm away. It is further to the side of and below the power socket, about 36 cm away." - }, - { - "id": "618310ed87_3", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"pot\", \"references\": [{\"object\": \"pot\", \"relation\": \"side_below\", \"distance_cm\": 28.4}, {\"object\": \"picture\", \"relation\": \"side_below\", \"distance_cm\": 33.1}]}" - }, - { - "id": "618310ed87_4", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power bank appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "618310ed87_5", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. dish sponge\nB. dish cloth\nC. cloth\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. cloth" - }, - { - "id": "618310ed87_6", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the shelf. It is also to the side of and above the cloth, about 5 cm away. It is further to the side of and above the heater, about 34 cm away." - }, - { - "id": "618310ed87_7", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"cloth\", \"relation\": \"side_above\", \"distance_cm\": 4.6}, {\"object\": \"heater\", \"relation\": \"side_above\", \"distance_cm\": 33.6}]}" - }, - { - "id": "618310ed87_8", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the scissors appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "618310ed87_9", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the handbag in this scene?\nA. power sockets\nB. table\nC. chair\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. table" - }, - { - "id": "618310ed87_10", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the handbag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The handbag is to the side of and below the table, about 6 cm away. It is also to the side of and below the power sockets, about 64 cm away. It is further near the power socket, about 71 cm away." - }, - { - "id": "618310ed87_11", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the handbag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"handbag\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 6.4}, {\"object\": \"power sockets\", \"relation\": \"side_below\", \"distance_cm\": 64.2}]}" - }, - { - "id": "618310ed87_12", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the smartphone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "618310ed87_13", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is to the side of and below the faucet, about 31 cm away. It is also near the dish sponge, about 31 cm away. It is further to the side of and above the dish cloth, about 50 cm away." - }, - { - "id": "618310ed87_14", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"sink\", \"references\": [{\"object\": \"faucet\", \"relation\": \"side_below\", \"distance_cm\": 30.6}, {\"object\": \"dish sponge\", \"relation\": \"near\", \"distance_cm\": 31.0}]}" - }, - { - "id": "618310ed87_15", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "618310ed87_16", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tv remote in this scene?\nA. pot\nB. picture\nC. tv\nD. backpack", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. tv" - }, - { - "id": "618310ed87_17", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tv remote in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tv remote is under the tv, about 3 cm away. It is also to the side of and below the picture, about 8 cm away. It is further to the side of and below the pot, about 51 cm away." - }, - { - "id": "618310ed87_18", - "dataset": "scannetpp", - "scene_name": "618310ed87", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tv remote in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tv remote\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"tv\", \"relation\": \"under\", \"distance_cm\": 2.6}, {\"object\": \"picture\", \"relation\": \"side_below\", \"distance_cm\": 7.8}]}" - }, - { - "id": "639f2c4d5a_0", - "dataset": "scannetpp", - "scene_name": "639f2c4d5a", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the extension cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "639f2c4d5a_1", - "dataset": "scannetpp", - "scene_name": "639f2c4d5a", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. faucet\nB. makeup pen\nC. soap dispenser\nD. shelf", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. makeup pen" - }, - { - "id": "639f2c4d5a_2", - "dataset": "scannetpp", - "scene_name": "639f2c4d5a", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is above the makeup pen. It is also next to the faucet, about 6 cm away. It is further to the side of and below the shelf, about 10 cm away." - }, - { - "id": "639f2c4d5a_3", - "dataset": "scannetpp", - "scene_name": "639f2c4d5a", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"makeup pen\", \"relation\": \"above\", \"distance_cm\": 2.3}, {\"object\": \"faucet\", \"relation\": \"next_to\", \"distance_cm\": 6.2}]}" - }, - { - "id": "64b8886d68_0", - "dataset": "scannetpp", - "scene_name": "64b8886d68", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "64b8886d68_1", - "dataset": "scannetpp", - "scene_name": "64b8886d68", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone charger in this scene?\nA. table\nB. light switch\nC. monitor\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. light switch" - }, - { - "id": "64b8886d68_2", - "dataset": "scannetpp", - "scene_name": "64b8886d68", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is to the side of and below the door, about 10 cm away. It is also below the light switch, about 76 cm away." - }, - { - "id": "64b8886d68_3", - "dataset": "scannetpp", - "scene_name": "64b8886d68", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"door\", \"references\": [{\"object\": \"door\", \"relation\": \"side_below\", \"distance_cm\": 10.0}, {\"object\": \"light switch\", \"relation\": \"below\", \"distance_cm\": 76.1}]}" - }, - { - "id": "666d04a14a_0", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power plug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "666d04a14a_1", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. doorframe\nB. shelf\nC. box\nD. apple", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. doorframe" - }, - { - "id": "666d04a14a_2", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the doorframe, about 5 cm away. It is also to the side of and above the door, about 5 cm away. It is further to the side of and below the shelf, about 24 cm away." - }, - { - "id": "666d04a14a_3", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"doorframe\", \"relation\": \"side_above\", \"distance_cm\": 4.7}, {\"object\": \"door\", \"relation\": \"side_above\", \"distance_cm\": 5.3}]}" - }, - { - "id": "666d04a14a_4", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the sticky note appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "666d04a14a_5", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. towel\nB. cloth\nC. book\nD. tray", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. tray" - }, - { - "id": "666d04a14a_6", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also next to the tray, about 3 cm away. It is further next to the cloth, about 9 cm away." - }, - { - "id": "666d04a14a_7", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"tray\", \"relation\": \"next_to\", \"distance_cm\": 3.1}, {\"object\": \"cloth\", \"relation\": \"next_to\", \"distance_cm\": 8.6}]}" - }, - { - "id": "666d04a14a_8", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "666d04a14a_9", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. document tray\nB. cabinet\nC. cloth\nD. desk orginizer", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cabinet" - }, - { - "id": "666d04a14a_10", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is on the table. It is also to the side of and below the cabinet, about 12 cm away. It is further to the side of and below the document tray, about 13 cm away." - }, - { - "id": "666d04a14a_11", - "dataset": "scannetpp", - "scene_name": "666d04a14a", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 11.6}, {\"object\": \"document tray\", \"relation\": \"side_below\", \"distance_cm\": 13.2}]}" - }, - { - "id": "67ab69f888_0", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the sticky note appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "67ab69f888_1", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the table. It is also next to the mug. It is further to the side of and below the monitor, about 7 cm away." - }, - { - "id": "67ab69f888_2", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 2.1}, {\"object\": \"mug\", \"relation\": \"next_to\", \"distance_cm\": 2.3}]}" - }, - { - "id": "67ab69f888_3", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "3", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "67ab69f888_4", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "4", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also next to the mousepad, about 3 cm away. It is further next to the mouse, about 4 cm away." - }, - { - "id": "67ab69f888_5", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "5", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"mousepad\", \"relation\": \"next_to\", \"distance_cm\": 3.4}, {\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 3.9}]}" - }, - { - "id": "67ab69f888_6", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "6", - "question_type": "target presence verification", - "question": "In the entire scene, did the adapter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "67ab69f888_7", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is next to the keyboard, about 4 cm away. It is also to the side of and above the table, about 4 cm away. It is further to the side of and below the monitor, about 20 cm away." - }, - { - "id": "67ab69f888_8", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 3.9}, {\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 4.0}]}" - }, - { - "id": "67ab69f888_9", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "9", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "67ab69f888_10", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "10", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mousepad in this scene?\nA. keyboard\nB. mug\nC. monitor\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. keyboard" - }, - { - "id": "67ab69f888_11", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "11", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mousepad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mousepad is on the table. It is also next to the keyboard, about 3 cm away. It is further to the side of and below the monitor, about 8 cm away." - }, - { - "id": "67ab69f888_12", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "12", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mousepad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mousepad\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 3.4}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 7.6}]}" - }, - { - "id": "67ab69f888_13", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "13", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "67ab69f888_14", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "14", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. telephone\nB. bottle\nC. mousepad\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. bottle" - }, - { - "id": "67ab69f888_15", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "15", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also next to the bottle. It is further to the side of and below the monitor, about 6 cm away." - }, - { - "id": "67ab69f888_16", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "16", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"bottle\", \"relation\": \"next_to\", \"distance_cm\": 2.3}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 6.5}]}" - }, - { - "id": "67ab69f888_17", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "17", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "67ab69f888_18", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "18", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen in this scene?\nA. bottle\nB. computer tower\nC. trash bin\nD. file binder", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. file binder" - }, - { - "id": "67ab69f888_19", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "19", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen is to the side of and below the file binder, about 10 cm away. It is also to the side of and above the trash bin, about 74 cm away." - }, - { - "id": "67ab69f888_20", - "dataset": "scannetpp", - "scene_name": "67ab69f888", - "sample_id": "20", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen\", \"support_surface\": \"file rack\", \"references\": [{\"object\": \"file binder\", \"relation\": \"side_below\", \"distance_cm\": 9.6}, {\"object\": \"trash bin\", \"relation\": \"side_above\", \"distance_cm\": 73.7}]}" - }, - { - "id": "6855e1ac32_0", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "6855e1ac32_1", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. shelf\nB. table\nC. kettle\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. table" - }, - { - "id": "6855e1ac32_2", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the table, about 15 cm away. It is also to the side of and below the shelf, about 32 cm away. It is further to the side of and below the kettle, about 43 cm away." - }, - { - "id": "6855e1ac32_3", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 15.1}, {\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 32.1}]}" - }, - { - "id": "6855e1ac32_4", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "6855e1ac32_5", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. sink\nB. tap\nC. split\nD. towel", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. tap" - }, - { - "id": "6855e1ac32_6", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the shelf. It is also to the side of and above the tap, about 8 cm away. It is further to the side of and above the sink, about 13 cm away." - }, - { - "id": "6855e1ac32_7", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"tap\", \"relation\": \"side_above\", \"distance_cm\": 8.2}, {\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 12.8}]}" - }, - { - "id": "6855e1ac32_8", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphone case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "6855e1ac32_9", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the purse in this scene?\nA. cabinet\nB. ottoman\nC. light switch\nD. jacket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. cabinet" - }, - { - "id": "6855e1ac32_10", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the purse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The purse is on the shelf. It is also to the side of and below the cabinet, about 9 cm away. It is further to the side of and below the jacket, about 35 cm away." - }, - { - "id": "6855e1ac32_11", - "dataset": "scannetpp", - "scene_name": "6855e1ac32", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the purse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"purse\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 8.5}, {\"object\": \"jacket\", \"relation\": \"side_below\", \"distance_cm\": 34.7}]}" - }, - { - "id": "689fec23d7_0", - "dataset": "scannetpp", - "scene_name": "689fec23d7", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "689fec23d7_1", - "dataset": "scannetpp", - "scene_name": "689fec23d7", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headset in this scene?\nA. office chair\nB. phone\nC. cup\nD. keyboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. phone" - }, - { - "id": "689fec23d7_2", - "dataset": "scannetpp", - "scene_name": "689fec23d7", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headset in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headset is on the desk. It is also next to the phone. It is further next to the cup, about 5 cm away." - }, - { - "id": "689fec23d7_3", - "dataset": "scannetpp", - "scene_name": "689fec23d7", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headset in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headset\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"phone\", \"relation\": \"next_to\", \"distance_cm\": 2.5}, {\"object\": \"cup\", \"relation\": \"next_to\", \"distance_cm\": 4.9}]}" - }, - { - "id": "689fec23d7_4", - "dataset": "scannetpp", - "scene_name": "689fec23d7", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "689fec23d7_5", - "dataset": "scannetpp", - "scene_name": "689fec23d7", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. keyboard\nB. cup\nC. office chair\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cup" - }, - { - "id": "689fec23d7_6", - "dataset": "scannetpp", - "scene_name": "689fec23d7", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also next to the cup, about 6 cm away. It is further next to the keyboard, about 7 cm away." - }, - { - "id": "689fec23d7_7", - "dataset": "scannetpp", - "scene_name": "689fec23d7", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"cup\", \"relation\": \"next_to\", \"distance_cm\": 5.5}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 7.3}]}" - }, - { - "id": "6c75a5290e_0", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the key appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "6c75a5290e_1", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. backpack\nB. blanket\nC. books\nD. bedside cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. blanket" - }, - { - "id": "6c75a5290e_2", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and below the blanket, about 33 cm away." - }, - { - "id": "6c75a5290e_3", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"blanket\", \"relation\": \"side_below\", \"distance_cm\": 32.6}]}" - }, - { - "id": "6c75a5290e_4", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote control appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "6c75a5290e_5", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen holder in this scene?\nA. tablet\nB. books\nC. phone\nD. table lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. tablet" - }, - { - "id": "6c75a5290e_6", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen holder in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen holder is on the cabinet. It is also to the side of and above the tablet, about 4 cm away. It is further next to the books, about 9 cm away." - }, - { - "id": "6c75a5290e_7", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen holder in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen holder\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"tablet\", \"relation\": \"side_above\", \"distance_cm\": 4.0}, {\"object\": \"books\", \"relation\": \"next_to\", \"distance_cm\": 9.4}]}" - }, - { - "id": "6c75a5290e_8", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "6c75a5290e_9", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone in this scene?\nA. table lamp\nB. tablet\nC. books\nD. pen holder", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. tablet" - }, - { - "id": "6c75a5290e_10", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone is on the cabinet. It is also next to the tablet. It is further to the side of and below the table lamp, about 12 cm away." - }, - { - "id": "6c75a5290e_11", - "dataset": "scannetpp", - "scene_name": "6c75a5290e", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"tablet\", \"relation\": \"next_to\", \"distance_cm\": 2.0}, {\"object\": \"table lamp\", \"relation\": \"side_below\", \"distance_cm\": 11.9}]}" - }, - { - "id": "6cc2231b9c_0", - "dataset": "scannetpp", - "scene_name": "6cc2231b9c", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the hairbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "6cc2231b9c_1", - "dataset": "scannetpp", - "scene_name": "6cc2231b9c", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen holder in this scene?\nA. table\nB. keyboard\nC. monitor\nD. telephone", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "6cc2231b9c_2", - "dataset": "scannetpp", - "scene_name": "6cc2231b9c", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen holder in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen holder is on the table. It is also to the side of and below the monitor, about 12 cm away. It is further near the keyboard, about 40 cm away." - }, - { - "id": "6cc2231b9c_3", - "dataset": "scannetpp", - "scene_name": "6cc2231b9c", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen holder in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen holder\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 11.6}, {\"object\": \"keyboard\", \"relation\": \"near\", \"distance_cm\": 39.8}]}" - }, - { - "id": "6e420fd87a_0", - "dataset": "scannetpp", - "scene_name": "6e420fd87a", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power plug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "6e420fd87a_1", - "dataset": "scannetpp", - "scene_name": "6e420fd87a", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. power socket\nB. outlet\nC. chair\nD. pillar", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. pillar" - }, - { - "id": "6e420fd87a_2", - "dataset": "scannetpp", - "scene_name": "6e420fd87a", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the pillar, about 27 cm away. It is also to the side of and above the chair, about 46 cm away. It is further near the power socket, about 72 cm away." - }, - { - "id": "6e420fd87a_3", - "dataset": "scannetpp", - "scene_name": "6e420fd87a", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"classroom table\", \"references\": [{\"object\": \"pillar\", \"relation\": \"side_below\", \"distance_cm\": 26.6}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 46.5}]}" - }, - { - "id": "6ef2ac745a_0", - "dataset": "scannetpp", - "scene_name": "6ef2ac745a", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "6ef2ac745a_1", - "dataset": "scannetpp", - "scene_name": "6ef2ac745a", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. power socket\nB. computer tower\nC. flipchart\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. power socket" - }, - { - "id": "6ef2ac745a_2", - "dataset": "scannetpp", - "scene_name": "6ef2ac745a", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the power socket, about 79 cm away." - }, - { - "id": "6ef2ac745a_3", - "dataset": "scannetpp", - "scene_name": "6ef2ac745a", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"flipchart\", \"references\": [{\"object\": \"power socket\", \"relation\": \"side_below\", \"distance_cm\": 79.1}]}" - }, - { - "id": "6ef2ac745a_4", - "dataset": "scannetpp", - "scene_name": "6ef2ac745a", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "6ef2ac745a_5", - "dataset": "scannetpp", - "scene_name": "6ef2ac745a", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. computer tower\nB. monitor\nC. chair\nD. office partition", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. monitor" - }, - { - "id": "6ef2ac745a_6", - "dataset": "scannetpp", - "scene_name": "6ef2ac745a", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also to the side of and below the monitor, about 11 cm away. It is further to the side of and above the chair, about 32 cm away." - }, - { - "id": "6ef2ac745a_7", - "dataset": "scannetpp", - "scene_name": "6ef2ac745a", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 10.9}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 32.3}]}" - }, - { - "id": "70945f435a_0", - "dataset": "scannetpp", - "scene_name": "70945f435a", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the stapler appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "70945f435a_1", - "dataset": "scannetpp", - "scene_name": "70945f435a", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. electric kettle\nB. refrigerator\nC. tray\nD. magnet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. tray" - }, - { - "id": "70945f435a_2", - "dataset": "scannetpp", - "scene_name": "70945f435a", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is under the tray, about 3 cm away. It is also to the side of and above the refrigerator, about 19 cm away. It is further to the side of and below the electric kettle, about 23 cm away." - }, - { - "id": "70945f435a_3", - "dataset": "scannetpp", - "scene_name": "70945f435a", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"tray\", \"references\": [{\"object\": \"tray\", \"relation\": \"under\", \"distance_cm\": 3.0}, {\"object\": \"refrigerator\", \"relation\": \"side_above\", \"distance_cm\": 19.0}]}" - }, - { - "id": "73f9370962_0", - "dataset": "scannetpp", - "scene_name": "73f9370962", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power bank appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "73f9370962_1", - "dataset": "scannetpp", - "scene_name": "73f9370962", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. monitor\nB. mouse\nC. windowsill\nD. paper", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. monitor" - }, - { - "id": "73f9370962_2", - "dataset": "scannetpp", - "scene_name": "73f9370962", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the cabinet. It is also to the side of and below the monitor, about 12 cm away. It is further to the side of and below the paper, about 16 cm away." - }, - { - "id": "73f9370962_3", - "dataset": "scannetpp", - "scene_name": "73f9370962", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 12.4}, {\"object\": \"paper\", \"relation\": \"side_below\", \"distance_cm\": 15.8}]}" - }, - { - "id": "73f9370962_4", - "dataset": "scannetpp", - "scene_name": "73f9370962", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "73f9370962_5", - "dataset": "scannetpp", - "scene_name": "73f9370962", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. paper\nB. windowsill\nC. monitor\nD. keyboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. monitor" - }, - { - "id": "73f9370962_6", - "dataset": "scannetpp", - "scene_name": "73f9370962", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the cabinet. It is also to the side of and below the monitor, about 26 cm away. It is further to the side of and below the paper, about 29 cm away." - }, - { - "id": "73f9370962_7", - "dataset": "scannetpp", - "scene_name": "73f9370962", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 26.2}, {\"object\": \"paper\", \"relation\": \"side_below\", \"distance_cm\": 29.2}]}" - }, - { - "id": "778769dad8_0", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "778769dad8_1", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. books\nB. paper bag\nC. laptop\nD. desk", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. desk" - }, - { - "id": "778769dad8_2", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the desk, about 16 cm away. It is also near the laptop, about 43 cm away. It is further to the side of and below the books, about 58 cm away." - }, - { - "id": "778769dad8_3", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"desk\", \"relation\": \"side_above\", \"distance_cm\": 16.1}, {\"object\": \"laptop\", \"relation\": \"near\", \"distance_cm\": 43.2}]}" - }, - { - "id": "778769dad8_4", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "778769dad8_5", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. kitchen utensils\nB. basket\nC. kitchen appliance\nD. jar", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. jar" - }, - { - "id": "778769dad8_6", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is next to the jar, about 5 cm away. It is also to the side of and above the kitchen appliance, about 24 cm away. It is further to the side of and below the basket, about 25 cm away." - }, - { - "id": "778769dad8_7", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"jar\", \"relation\": \"next_to\", \"distance_cm\": 4.9}, {\"object\": \"kitchen appliance\", \"relation\": \"side_above\", \"distance_cm\": 23.9}]}" - }, - { - "id": "778769dad8_8", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue box appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "778769dad8_9", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the desk. It is also next to the books, about 5 cm away. It is further under the bottle, about 5 cm away." - }, - { - "id": "778769dad8_10", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"books\", \"relation\": \"next_to\", \"distance_cm\": 4.7}, {\"object\": \"bottle\", \"relation\": \"under\", \"distance_cm\": 4.8}]}" - }, - { - "id": "778769dad8_11", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the headset appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "778769dad8_12", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone stand in this scene?\nA. chair\nB. books\nC. headphones\nD. laptop", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. laptop" - }, - { - "id": "778769dad8_13", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone stand in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone stand is on the desk. It is also next to the laptop, about 13 cm away. It is further near the books, about 17 cm away." - }, - { - "id": "778769dad8_14", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone stand in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone stand\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 13.2}, {\"object\": \"books\", \"relation\": \"near\", \"distance_cm\": 17.4}]}" - }, - { - "id": "778769dad8_15", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "778769dad8_16", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. pillow\nB. mattress\nC. heater\nD. electric kettle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. electric kettle" - }, - { - "id": "778769dad8_17", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is next to the electric kettle, about 7 cm away. It is also to the side of and below the mattress, about 24 cm away. It is further to the side of and below the heater, about 31 cm away." - }, - { - "id": "778769dad8_18", - "dataset": "scannetpp", - "scene_name": "778769dad8", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"mattress\", \"references\": [{\"object\": \"electric kettle\", \"relation\": \"next_to\", \"distance_cm\": 7.0}, {\"object\": \"mattress\", \"relation\": \"side_below\", \"distance_cm\": 23.6}]}" - }, - { - "id": "7968e15e5e_0", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the highlighter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7968e15e5e_1", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the marker in this scene?\nA. door frame\nB. storage cabinet\nC. backpack\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. storage cabinet" - }, - { - "id": "7968e15e5e_2", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the marker in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The marker is near the storage cabinet, about 66 cm away. It is also to the side of and below the door, about 71 cm away. It is further to the side of and below the door frame, about 76 cm away." - }, - { - "id": "7968e15e5e_3", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the marker in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"marker\", \"support_surface\": \"storage cabinet\", \"references\": [{\"object\": \"storage cabinet\", \"relation\": \"near\", \"distance_cm\": 65.6}, {\"object\": \"door\", \"relation\": \"side_below\", \"distance_cm\": 70.9}]}" - }, - { - "id": "7968e15e5e_4", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7968e15e5e_5", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse pad in this scene?\nA. sticky note\nB. usb hub\nC. file cabinet\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. sticky note" - }, - { - "id": "7968e15e5e_6", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse pad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse pad is on the table. It is also next to the sticky note, about 12 cm away. It is further to the side of and below the monitor, about 19 cm away." - }, - { - "id": "7968e15e5e_7", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse pad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse pad\", \"support_surface\": \"table\", \"references\": [{\"object\": \"sticky note\", \"relation\": \"next_to\", \"distance_cm\": 11.7}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 19.3}]}" - }, - { - "id": "7968e15e5e_8", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the usb hub appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "7968e15e5e_9", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the usb hub in this scene?\nA. hand sanitizer\nB. sticky note\nC. mouse pad\nD. laptop", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. sticky note" - }, - { - "id": "7968e15e5e_10", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the usb hub in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The usb hub is on the table. It is also next to the sticky note, about 5 cm away. It is further to the side of and below the laptop, about 7 cm away." - }, - { - "id": "7968e15e5e_11", - "dataset": "scannetpp", - "scene_name": "7968e15e5e", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the usb hub in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"usb hub\", \"support_surface\": \"table\", \"references\": [{\"object\": \"sticky note\", \"relation\": \"next_to\", \"distance_cm\": 5.3}, {\"object\": \"laptop\", \"relation\": \"side_below\", \"distance_cm\": 6.6}]}" - }, - { - "id": "7a629f789c_0", - "dataset": "scannetpp", - "scene_name": "7a629f789c", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7a629f789c_1", - "dataset": "scannetpp", - "scene_name": "7a629f789c", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. trash can\nB. pillow\nC. shelf\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. trash can" - }, - { - "id": "7a629f789c_2", - "dataset": "scannetpp", - "scene_name": "7a629f789c", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the nightstand. It is also to the side of and above the trash can, about 19 cm away. It is further near the cup, about 31 cm away." - }, - { - "id": "7a629f789c_3", - "dataset": "scannetpp", - "scene_name": "7a629f789c", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"nightstand\", \"references\": [{\"object\": \"trash can\", \"relation\": \"side_above\", \"distance_cm\": 18.8}, {\"object\": \"cup\", \"relation\": \"near\", \"distance_cm\": 30.7}]}" - }, - { - "id": "7b04052ad0_0", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7b04052ad0_1", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. box\nB. power strip\nC. dustbin\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. box" - }, - { - "id": "7b04052ad0_2", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the rug, about 3 cm away. It is also to the side of and above the power strip, about 16 cm away. It is further to the side of and below the table, about 19 cm away." - }, - { - "id": "7b04052ad0_3", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"rug\", \"relation\": \"side_above\", \"distance_cm\": 2.6}, {\"object\": \"power strip\", \"relation\": \"side_above\", \"distance_cm\": 16.0}]}" - }, - { - "id": "7b04052ad0_4", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7b04052ad0_5", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. bottle\nB. box\nC. paper\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. paper" - }, - { - "id": "7b04052ad0_6", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is to the side of and below the paper, about 7 cm away. It is also to the side of and above the box, about 27 cm away. It is further to the side of and below the cabinet, about 29 cm away." - }, - { - "id": "7b04052ad0_7", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"side table\", \"references\": [{\"object\": \"paper\", \"relation\": \"side_below\", \"distance_cm\": 6.8}, {\"object\": \"box\", \"relation\": \"side_above\", \"distance_cm\": 26.9}]}" - }, - { - "id": "7b04052ad0_8", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "7b04052ad0_9", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. chair\nB. laptop\nC. cloth\nD. toilet paper roll", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. laptop" - }, - { - "id": "7b04052ad0_10", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also next to the laptop, about 12 cm away. It is further to the side of and below the cloth, about 14 cm away." - }, - { - "id": "7b04052ad0_11", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 12.0}, {\"object\": \"cloth\", \"relation\": \"side_below\", \"distance_cm\": 13.8}]}" - }, - { - "id": "7b04052ad0_12", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "7b04052ad0_13", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. satchel\nB. box\nC. backpack\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. heater" - }, - { - "id": "7b04052ad0_14", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is on the rug. It is also to the side of and below the heater, about 10 cm away. It is further to the side of and below the backpack, about 16 cm away." - }, - { - "id": "7b04052ad0_15", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"rug\", \"references\": [{\"object\": \"heater\", \"relation\": \"side_below\", \"distance_cm\": 9.6}, {\"object\": \"backpack\", \"relation\": \"side_below\", \"distance_cm\": 16.0}]}" - }, - { - "id": "7b04052ad0_16", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the tape dispenser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7b04052ad0_17", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "17", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the soap in this scene?\nA. air fryer\nB. window sill\nC. dish sponge\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bottle" - }, - { - "id": "7b04052ad0_18", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "18", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap is to the side of and below the bottle, about 43 cm away. It is also to the side of and below the window sill, about 58 cm away. It is further to the side of and below the air fryer, about 70 cm away." - }, - { - "id": "7b04052ad0_19", - "dataset": "scannetpp", - "scene_name": "7b04052ad0", - "sample_id": "19", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap\", \"support_surface\": \"window sill\", \"references\": [{\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 43.4}, {\"object\": \"window sill\", \"relation\": \"side_below\", \"distance_cm\": 58.3}]}" - }, - { - "id": "7bc286c1b6_0", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the notepad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7bc286c1b6_1", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. sink\nB. razor\nC. shaving foam\nD. faucet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. sink" - }, - { - "id": "7bc286c1b6_2", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the sink, about 4 cm away. It is also next to the faucet, about 7 cm away. It is further to the side of and below the shaving foam, about 15 cm away." - }, - { - "id": "7bc286c1b6_3", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"bathroom counter\", \"references\": [{\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 3.6}, {\"object\": \"faucet\", \"relation\": \"next_to\", \"distance_cm\": 6.6}]}" - }, - { - "id": "7bc286c1b6_4", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the razor appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "7bc286c1b6_5", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the razor in this scene?\nA. cloth\nB. faucet\nC. shower wall\nD. hand shower", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. faucet" - }, - { - "id": "7bc286c1b6_6", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the razor in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The razor is to the side of and below the faucet, about 6 cm away. It is also next to the cloth, about 13 cm away. It is further to the side of and below the shower wall, about 13 cm away." - }, - { - "id": "7bc286c1b6_7", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the razor in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"razor\", \"support_surface\": \"bathroom counter\", \"references\": [{\"object\": \"faucet\", \"relation\": \"side_below\", \"distance_cm\": 6.1}, {\"object\": \"cloth\", \"relation\": \"next_to\", \"distance_cm\": 13.0}]}" - }, - { - "id": "7bc286c1b6_8", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the game controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7bc286c1b6_9", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toothbrush in this scene?\nA. towel\nB. toilet flush button\nC. bathroom counter\nD. recessed shelve", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. recessed shelve" - }, - { - "id": "7bc286c1b6_10", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toothbrush in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toothbrush is below the recessed shelve, about 11 cm away. It is also to the side of and below the towel, about 39 cm away. It is further to the side of and above the bathroom counter, about 41 cm away." - }, - { - "id": "7bc286c1b6_11", - "dataset": "scannetpp", - "scene_name": "7bc286c1b6", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toothbrush in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toothbrush\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"recessed shelve\", \"relation\": \"below\", \"distance_cm\": 10.8}, {\"object\": \"towel\", \"relation\": \"side_below\", \"distance_cm\": 39.3}]}" - }, - { - "id": "7c31a42404_0", - "dataset": "scannetpp", - "scene_name": "7c31a42404", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cable appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7c31a42404_1", - "dataset": "scannetpp", - "scene_name": "7c31a42404", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tv remote in this scene?\nA. table\nB. couch\nC. chair\nD. ceiling lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "7c31a42404_2", - "dataset": "scannetpp", - "scene_name": "7c31a42404", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tv remote in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tv remote is to the side of and above the table, about 7 cm away. It is also to the side of and above the chair, about 41 cm away. It is further to the side of and below the ceiling lamp, about 50 cm away." - }, - { - "id": "7c31a42404_3", - "dataset": "scannetpp", - "scene_name": "7c31a42404", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tv remote in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tv remote\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 7.1}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 41.4}]}" - }, - { - "id": "7f77abce34_0", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the game controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7f77abce34_1", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. kitchen cabinet\nB. oven\nC. jar\nD. stove", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. jar" - }, - { - "id": "7f77abce34_2", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is near the jar, about 16 cm away. It is also to the side of and above the stove, about 32 cm away. It is further to the side of and above the oven, about 49 cm away." - }, - { - "id": "7f77abce34_3", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"jar\", \"relation\": \"near\", \"distance_cm\": 15.8}, {\"object\": \"stove\", \"relation\": \"side_above\", \"distance_cm\": 32.1}]}" - }, - { - "id": "7f77abce34_4", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "7f77abce34_5", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also next to the laptop, about 14 cm away. It is further to the side of and below the wardrobe, about 14 cm away." - }, - { - "id": "7f77abce34_6", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 13.8}, {\"object\": \"wardrobe\", \"relation\": \"side_below\", \"distance_cm\": 14.2}]}" - }, - { - "id": "7f77abce34_7", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the hairbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7f77abce34_8", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the notebook in this scene?\nA. monitor\nB. laptop\nC. table\nD. office chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. table" - }, - { - "id": "7f77abce34_9", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the notebook in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The notebook is to the side of and above the table, about 5 cm away. It is also to the side of and above the laptop, about 13 cm away. It is further to the side of and above the office chair, about 34 cm away." - }, - { - "id": "7f77abce34_10", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the notebook in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"notebook\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 4.9}, {\"object\": \"laptop\", \"relation\": \"side_above\", \"distance_cm\": 13.0}]}" - }, - { - "id": "7f77abce34_11", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "7f77abce34_12", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. wardrobe\nB. mouse\nC. split\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. monitor" - }, - { - "id": "7f77abce34_13", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is on the table. It is also to the side of and below the monitor, about 12 cm away. It is further near the split, about 30 cm away." - }, - { - "id": "7f77abce34_14", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 11.9}, {\"object\": \"split\", \"relation\": \"near\", \"distance_cm\": 29.9}]}" - }, - { - "id": "7f77abce34_15", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the tote bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "7f77abce34_16", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tote bag in this scene?\nA. office chair\nB. backpack\nC. light switch\nD. doorframe", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. office chair" - }, - { - "id": "7f77abce34_17", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tote bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tote bag is on the desk. It is also to the side of and above the office chair, about 14 cm away. It is further near the backpack, about 17 cm away." - }, - { - "id": "7f77abce34_18", - "dataset": "scannetpp", - "scene_name": "7f77abce34", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tote bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tote bag\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 14.4}, {\"object\": \"backpack\", \"relation\": \"near\", \"distance_cm\": 16.9}]}" - }, - { - "id": "8890d0a267_0", - "dataset": "scannetpp", - "scene_name": "8890d0a267", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the eyeglasses appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "8890d0a267_1", - "dataset": "scannetpp", - "scene_name": "8890d0a267", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. bed\nB. blanket\nC. bottle\nD. bag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. bottle" - }, - { - "id": "8890d0a267_2", - "dataset": "scannetpp", - "scene_name": "8890d0a267", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the bottle, about 28 cm away." - }, - { - "id": "8890d0a267_3", - "dataset": "scannetpp", - "scene_name": "8890d0a267", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 27.6}]}" - }, - { - "id": "891865308a_0", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cable appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "891865308a_1", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the extension cord in this scene?\nA. power socket\nB. bed\nC. folded blanket\nD. blanket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. folded blanket" - }, - { - "id": "891865308a_2", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the extension cord in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The extension cord is to the side of and below the folded blanket, about 50 cm away. It is also to the side of and below the power socket, about 72 cm away." - }, - { - "id": "891865308a_3", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the extension cord in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"extension cord\", \"support_surface\": \"bed\", \"references\": [{\"object\": \"folded blanket\", \"relation\": \"side_below\", \"distance_cm\": 50.4}, {\"object\": \"power socket\", \"relation\": \"side_below\", \"distance_cm\": 72.4}]}" - }, - { - "id": "891865308a_4", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "891865308a_5", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. mouse\nB. lotion bottle\nC. chair\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. monitor" - }, - { - "id": "891865308a_6", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the desk. It is also to the side of and below the monitor, about 8 cm away. It is further near the mouse, about 15 cm away." - }, - { - "id": "891865308a_7", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 8.5}, {\"object\": \"mouse\", \"relation\": \"near\", \"distance_cm\": 15.2}]}" - }, - { - "id": "891865308a_8", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the purse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "891865308a_9", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the lotion bottle in this scene?\nA. mouse\nB. headphones\nC. monitor\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. headphones" - }, - { - "id": "891865308a_10", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the lotion bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The lotion bottle is on the desk. It is also near the headphones, about 25 cm away. It is further near the mouse, about 35 cm away." - }, - { - "id": "891865308a_11", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the lotion bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"lotion bottle\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"headphones\", \"relation\": \"near\", \"distance_cm\": 25.1}, {\"object\": \"mouse\", \"relation\": \"near\", \"distance_cm\": 35.1}]}" - }, - { - "id": "891865308a_12", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "891865308a_13", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. monitor\nB. headphones\nC. chair\nD. lotion bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. headphones" - }, - { - "id": "891865308a_14", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also near the headphones, about 15 cm away. It is further to the side of and above the chair, about 23 cm away." - }, - { - "id": "891865308a_15", - "dataset": "scannetpp", - "scene_name": "891865308a", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"headphones\", \"relation\": \"near\", \"distance_cm\": 15.2}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 23.4}]}" - }, - { - "id": "893fb90e89_0", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "893fb90e89_1", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. binder\nB. bag\nC. books\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. table" - }, - { - "id": "893fb90e89_2", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the table, about 18 cm away." - }, - { - "id": "893fb90e89_3", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 17.7}]}" - }, - { - "id": "893fb90e89_4", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the notebook appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "893fb90e89_5", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the notebook in this scene?\nA. pencil\nB. split\nC. office chair\nD. tray", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. pencil" - }, - { - "id": "893fb90e89_6", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the notebook in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The notebook is on the table. It is also near the pencil, about 16 cm away. It is further to the side of and above the office chair, about 37 cm away." - }, - { - "id": "893fb90e89_7", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the notebook in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"notebook\", \"support_surface\": \"table\", \"references\": [{\"object\": \"pencil\", \"relation\": \"near\", \"distance_cm\": 16.2}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 36.6}]}" - }, - { - "id": "893fb90e89_8", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "893fb90e89_9", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen holder in this scene?\nA. poster\nB. table\nC. box\nD. file holder", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. box" - }, - { - "id": "893fb90e89_10", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen holder in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen holder is on the cabinet. It is also next to the box, about 6 cm away. It is further to the side of and above the table, about 11 cm away." - }, - { - "id": "893fb90e89_11", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen holder in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen holder\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 5.9}, {\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 11.1}]}" - }, - { - "id": "893fb90e89_12", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "893fb90e89_13", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pencil in this scene?\nA. notebook\nB. office chair\nC. split\nD. tray", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. notebook" - }, - { - "id": "893fb90e89_14", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pencil in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pencil is on the table. It is also near the notebook, about 16 cm away. It is further near the split, about 28 cm away." - }, - { - "id": "893fb90e89_15", - "dataset": "scannetpp", - "scene_name": "893fb90e89", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pencil in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pencil\", \"support_surface\": \"table\", \"references\": [{\"object\": \"notebook\", \"relation\": \"near\", \"distance_cm\": 16.2}, {\"object\": \"split\", \"relation\": \"near\", \"distance_cm\": 28.3}]}" - }, - { - "id": "8b5caf3398_0", - "dataset": "scannetpp", - "scene_name": "8b5caf3398", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "8b5caf3398_1", - "dataset": "scannetpp", - "scene_name": "8b5caf3398", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. paper\nB. window frame\nC. cabinet\nD. tissue box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. window frame" - }, - { - "id": "8b5caf3398_2", - "dataset": "scannetpp", - "scene_name": "8b5caf3398", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and below the window frame, about 8 cm away. It is also to the side of and below the window, about 32 cm away. It is further to the side of and above the tissue box, about 83 cm away." - }, - { - "id": "8b5caf3398_3", - "dataset": "scannetpp", - "scene_name": "8b5caf3398", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"window frame\", \"references\": [{\"object\": \"window frame\", \"relation\": \"side_below\", \"distance_cm\": 8.0}, {\"object\": \"window\", \"relation\": \"side_below\", \"distance_cm\": 31.5}]}" - }, - { - "id": "8b5caf3398_4", - "dataset": "scannetpp", - "scene_name": "8b5caf3398", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the wallet appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "8b5caf3398_5", - "dataset": "scannetpp", - "scene_name": "8b5caf3398", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue box in this scene?\nA. cabinet\nB. bottle\nC. window frame\nD. paper", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. paper" - }, - { - "id": "8b5caf3398_6", - "dataset": "scannetpp", - "scene_name": "8b5caf3398", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue box in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue box is on the cabinet. It is also to the side of and above the paper, about 5 cm away. It is further to the side of and below the bottle, about 83 cm away." - }, - { - "id": "8b5caf3398_7", - "dataset": "scannetpp", - "scene_name": "8b5caf3398", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue box in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue box\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"paper\", \"relation\": \"side_above\", \"distance_cm\": 4.7}, {\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 82.8}]}" - }, - { - "id": "8be0cd3817_0", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "8be0cd3817_1", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. pan\nB. kitchen counter\nC. dish rack\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. power socket" - }, - { - "id": "8be0cd3817_2", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the shelf. It is also to the side of and above the power socket, about 28 cm away. It is further to the side of and above the pan, about 37 cm away." - }, - { - "id": "8be0cd3817_3", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"power socket\", \"relation\": \"side_above\", \"distance_cm\": 27.8}, {\"object\": \"pan\", \"relation\": \"side_above\", \"distance_cm\": 37.1}]}" - }, - { - "id": "8be0cd3817_4", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "8be0cd3817_5", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the earbuds in this scene?\nA. blanket\nB. laptop\nC. shelf\nD. table lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. shelf" - }, - { - "id": "8be0cd3817_6", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the earbuds in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The earbuds is above the shelf, about 3 cm away. It is also next to the laptop, about 5 cm away. It is further to the side of and below the table lamp, about 8 cm away." - }, - { - "id": "8be0cd3817_7", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the earbuds in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"earbuds\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"shelf\", \"relation\": \"above\", \"distance_cm\": 2.8}, {\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 5.4}]}" - }, - { - "id": "8be0cd3817_8", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "8be0cd3817_9", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. power strip\nB. wardrobe\nC. office chair\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. monitor" - }, - { - "id": "8be0cd3817_10", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also to the side of and below the monitor, about 6 cm away. It is further to the side of and above the office chair, about 21 cm away." - }, - { - "id": "8be0cd3817_11", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 5.7}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 21.2}]}" - }, - { - "id": "8be0cd3817_12", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the power cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "8be0cd3817_13", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. keyboard\nB. monitor\nC. kettle\nD. toaster", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. monitor" - }, - { - "id": "8be0cd3817_14", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is on the table. It is also to the side of and below the monitor, about 14 cm away. It is further to the side of and below the kettle, about 26 cm away." - }, - { - "id": "8be0cd3817_15", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 13.8}, {\"object\": \"kettle\", \"relation\": \"side_below\", \"distance_cm\": 26.2}]}" - }, - { - "id": "8be0cd3817_16", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the deodorant bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "8be0cd3817_17", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the remote control in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The remote control is next to the powerbank, about 8 cm away. It is also to the side of and below the blanket, about 8 cm away. It is further near the dvd player, about 18 cm away." - }, - { - "id": "8be0cd3817_18", - "dataset": "scannetpp", - "scene_name": "8be0cd3817", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the remote control in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"remote control\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"powerbank\", \"relation\": \"next_to\", \"distance_cm\": 8.3}, {\"object\": \"blanket\", \"relation\": \"side_below\", \"distance_cm\": 8.3}]}" - }, - { - "id": "8fc40ba77b_0", - "dataset": "scannetpp", - "scene_name": "8fc40ba77b", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone stand appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "8fc40ba77b_1", - "dataset": "scannetpp", - "scene_name": "8fc40ba77b", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. windowsill\nB. heater\nC. chair\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. heater" - }, - { - "id": "8fc40ba77b_2", - "dataset": "scannetpp", - "scene_name": "8fc40ba77b", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the heater, about 11 cm away. It is also to the side of and below the power socket, about 19 cm away. It is further to the side of and below the windowsill, about 64 cm away." - }, - { - "id": "8fc40ba77b_3", - "dataset": "scannetpp", - "scene_name": "8fc40ba77b", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"heater\", \"relation\": \"side_below\", \"distance_cm\": 10.7}, {\"object\": \"power socket\", \"relation\": \"side_below\", \"distance_cm\": 18.6}]}" - }, - { - "id": "927aacd5d1_0", - "dataset": "scannetpp", - "scene_name": "927aacd5d1", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "927aacd5d1_1", - "dataset": "scannetpp", - "scene_name": "927aacd5d1", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. power socket\nB. box\nC. pillow\nD. table lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. table lamp" - }, - { - "id": "927aacd5d1_2", - "dataset": "scannetpp", - "scene_name": "927aacd5d1", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the nightstand. It is also to the side of and below the table lamp, about 7 cm away. It is further next to the power socket, about 10 cm away." - }, - { - "id": "927aacd5d1_3", - "dataset": "scannetpp", - "scene_name": "927aacd5d1", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"nightstand\", \"references\": [{\"object\": \"table lamp\", \"relation\": \"side_below\", \"distance_cm\": 6.5}, {\"object\": \"power socket\", \"relation\": \"next_to\", \"distance_cm\": 10.1}]}" - }, - { - "id": "927aacd5d1_4", - "dataset": "scannetpp", - "scene_name": "927aacd5d1", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the stapler appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "927aacd5d1_5", - "dataset": "scannetpp", - "scene_name": "927aacd5d1", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the spray bottle in this scene?\nA. webcam\nB. cap\nC. box\nD. clothes", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cap" - }, - { - "id": "927aacd5d1_6", - "dataset": "scannetpp", - "scene_name": "927aacd5d1", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is on the table. It is also next to the cap. It is further near the webcam, about 31 cm away." - }, - { - "id": "927aacd5d1_7", - "dataset": "scannetpp", - "scene_name": "927aacd5d1", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"cap\", \"relation\": \"next_to\", \"distance_cm\": 2.4}, {\"object\": \"webcam\", \"relation\": \"near\", \"distance_cm\": 31.0}]}" - }, - { - "id": "96b7c379eb_0", - "dataset": "scannetpp", - "scene_name": "96b7c379eb", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the usb hub appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "96b7c379eb_1", - "dataset": "scannetpp", - "scene_name": "96b7c379eb", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the spray bottle in this scene?\nA. toilet flush\nB. toilet\nC. power sockets\nD. shelf", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. toilet flush" - }, - { - "id": "96b7c379eb_2", - "dataset": "scannetpp", - "scene_name": "96b7c379eb", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is above the toilet flush, about 9 cm away. It is also to the side of and above the power sockets, about 25 cm away. It is further near the shelf, about 26 cm away." - }, - { - "id": "96b7c379eb_3", - "dataset": "scannetpp", - "scene_name": "96b7c379eb", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"toilet flush\", \"relation\": \"above\", \"distance_cm\": 8.9}, {\"object\": \"power sockets\", \"relation\": \"side_above\", \"distance_cm\": 24.8}]}" - }, - { - "id": "9816c49e97_0", - "dataset": "scannetpp", - "scene_name": "9816c49e97", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the soap appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "9816c49e97_1", - "dataset": "scannetpp", - "scene_name": "9816c49e97", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the water bottle in this scene?\nA. chair\nB. table\nC. monitor\nD. bag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. table" - }, - { - "id": "9816c49e97_2", - "dataset": "scannetpp", - "scene_name": "9816c49e97", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water bottle is to the side of and above the table, about 8 cm away. It is also to the side of and above the chair, about 16 cm away." - }, - { - "id": "9816c49e97_3", - "dataset": "scannetpp", - "scene_name": "9816c49e97", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 8.3}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 16.2}]}" - }, - { - "id": "99a2503d63_0", - "dataset": "scannetpp", - "scene_name": "99a2503d63", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "99a2503d63_1", - "dataset": "scannetpp", - "scene_name": "99a2503d63", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. chair\nB. microwave\nC. table\nD. bowl", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "99a2503d63_2", - "dataset": "scannetpp", - "scene_name": "99a2503d63", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the chair, about 5 cm away. It is also near the table, about 22 cm away. It is further to the side of and below the microwave, about 28 cm away." - }, - { - "id": "99a2503d63_3", - "dataset": "scannetpp", - "scene_name": "99a2503d63", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 5.1}, {\"object\": \"table\", \"relation\": \"near\", \"distance_cm\": 22.0}]}" - }, - { - "id": "99fa5c25e1_0", - "dataset": "scannetpp", - "scene_name": "99fa5c25e1", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cable appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "99fa5c25e1_1", - "dataset": "scannetpp", - "scene_name": "99fa5c25e1", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the notebook in this scene?\nA. hammer holder\nB. papers\nC. cloth\nD. paper holder", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. papers" - }, - { - "id": "99fa5c25e1_2", - "dataset": "scannetpp", - "scene_name": "99fa5c25e1", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the notebook in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The notebook is next to the papers, about 12 cm away. It is also near the cloth, about 16 cm away. It is further near the hammer holder, about 20 cm away." - }, - { - "id": "99fa5c25e1_3", - "dataset": "scannetpp", - "scene_name": "99fa5c25e1", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the notebook in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"notebook\", \"support_surface\": \"papers\", \"references\": [{\"object\": \"papers\", \"relation\": \"next_to\", \"distance_cm\": 12.2}, {\"object\": \"cloth\", \"relation\": \"near\", \"distance_cm\": 16.0}]}" - }, - { - "id": "9e9d017ad1_0", - "dataset": "scannetpp", - "scene_name": "9e9d017ad1", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "9e9d017ad1_1", - "dataset": "scannetpp", - "scene_name": "9e9d017ad1", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. sink\nB. soap dispenser\nC. cabinet\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. sink" - }, - { - "id": "9e9d017ad1_2", - "dataset": "scannetpp", - "scene_name": "9e9d017ad1", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the cabinet. It is also next to the sink, about 5 cm away. It is further to the side of and below the soap dispenser, about 63 cm away." - }, - { - "id": "9e9d017ad1_3", - "dataset": "scannetpp", - "scene_name": "9e9d017ad1", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"sink\", \"relation\": \"next_to\", \"distance_cm\": 4.9}, {\"object\": \"soap dispenser\", \"relation\": \"side_below\", \"distance_cm\": 63.1}]}" - }, - { - "id": "9f139a318d_0", - "dataset": "scannetpp", - "scene_name": "9f139a318d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "9f139a318d_1", - "dataset": "scannetpp", - "scene_name": "9f139a318d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. book\nB. keyboard\nC. telephone\nD. paper", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. keyboard" - }, - { - "id": "9f139a318d_2", - "dataset": "scannetpp", - "scene_name": "9f139a318d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also next to the keyboard, about 5 cm away. It is further next to the telephone, about 7 cm away." - }, - { - "id": "9f139a318d_3", - "dataset": "scannetpp", - "scene_name": "9f139a318d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 4.6}, {\"object\": \"telephone\", \"relation\": \"next_to\", \"distance_cm\": 7.1}]}" - }, - { - "id": "9f139a318d_4", - "dataset": "scannetpp", - "scene_name": "9f139a318d", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the highlighter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "9f139a318d_5", - "dataset": "scannetpp", - "scene_name": "9f139a318d", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. cabinet\nB. backpack\nC. table\nD. jacket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. cabinet" - }, - { - "id": "9f139a318d_6", - "dataset": "scannetpp", - "scene_name": "9f139a318d", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the cabinet, about 7 cm away. It is also to the side of and below the jacket, about 46 cm away. It is further to the side of and below the backpack, about 56 cm away." - }, - { - "id": "9f139a318d_7", - "dataset": "scannetpp", - "scene_name": "9f139a318d", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 7.0}, {\"object\": \"jacket\", \"relation\": \"side_below\", \"distance_cm\": 46.5}]}" - }, - { - "id": "a1d9da703c_0", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a1d9da703c_1", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. pen\nB. monitor\nC. box\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. monitor" - }, - { - "id": "a1d9da703c_2", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also under the monitor, about 4 cm away. It is further next to the box, about 6 cm away." - }, - { - "id": "a1d9da703c_3", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"under\", \"distance_cm\": 3.9}, {\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 6.2}]}" - }, - { - "id": "a1d9da703c_4", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a1d9da703c_5", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen in this scene?\nA. mouse\nB. chair\nC. mug\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. chair" - }, - { - "id": "a1d9da703c_6", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen is on the table. It is also to the side of and below the chair, about 10 cm away. It is further next to the mouse, about 12 cm away." - }, - { - "id": "a1d9da703c_7", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 9.8}, {\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 12.4}]}" - }, - { - "id": "a1d9da703c_8", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the stapler appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "a1d9da703c_9", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the stapler in this scene?\nA. box\nB. mug\nC. keyboard\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. box" - }, - { - "id": "a1d9da703c_10", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the stapler in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The stapler is on the table. It is also near the box, about 35 cm away. It is further near the keyboard, about 38 cm away." - }, - { - "id": "a1d9da703c_11", - "dataset": "scannetpp", - "scene_name": "a1d9da703c", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the stapler in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"stapler\", \"support_surface\": \"table\", \"references\": [{\"object\": \"box\", \"relation\": \"near\", \"distance_cm\": 35.2}, {\"object\": \"keyboard\", \"relation\": \"near\", \"distance_cm\": 38.0}]}" - }, - { - "id": "a30b666f83_0", - "dataset": "scannetpp", - "scene_name": "a30b666f83", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a30b666f83_1", - "dataset": "scannetpp", - "scene_name": "a30b666f83", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is to the side of and above the basket, about 3 cm away. It is also next to the dish soap bottle, about 3 cm away. It is further to the side of and below the power outlet, about 22 cm away." - }, - { - "id": "a30b666f83_2", - "dataset": "scannetpp", - "scene_name": "a30b666f83", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"basket\", \"references\": [{\"object\": \"basket\", \"relation\": \"side_above\", \"distance_cm\": 2.6}, {\"object\": \"dish soap bottle\", \"relation\": \"next_to\", \"distance_cm\": 3.4}]}" - }, - { - "id": "a492fe77aa_0", - "dataset": "scannetpp", - "scene_name": "a492fe77aa", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue box appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a492fe77aa_1", - "dataset": "scannetpp", - "scene_name": "a492fe77aa", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. compact wardrobe\nB. paper bag\nC. clothes\nD. picture frame", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. clothes" - }, - { - "id": "a492fe77aa_2", - "dataset": "scannetpp", - "scene_name": "a492fe77aa", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the picture frame, about 11 cm away. It is also to the side of and below the paper bag, about 39 cm away." - }, - { - "id": "a492fe77aa_3", - "dataset": "scannetpp", - "scene_name": "a492fe77aa", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"paper bag\", \"references\": [{\"object\": \"picture frame\", \"relation\": \"side_below\", \"distance_cm\": 10.8}, {\"object\": \"paper bag\", \"relation\": \"side_below\", \"distance_cm\": 39.4}]}" - }, - { - "id": "a492fe77aa_4", - "dataset": "scannetpp", - "scene_name": "a492fe77aa", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the extension cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "a492fe77aa_5", - "dataset": "scannetpp", - "scene_name": "a492fe77aa", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the extension cord in this scene?\nA. clothes\nB. backpack\nC. table\nD. plant pot", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. table" - }, - { - "id": "a492fe77aa_6", - "dataset": "scannetpp", - "scene_name": "a492fe77aa", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the extension cord in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The extension cord is to the side of and below the table, about 23 cm away. It is also to the side of and below the plant pot, about 38 cm away." - }, - { - "id": "a492fe77aa_7", - "dataset": "scannetpp", - "scene_name": "a492fe77aa", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the extension cord in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"extension cord\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 23.4}, {\"object\": \"plant pot\", \"relation\": \"side_below\", \"distance_cm\": 38.4}]}" - }, - { - "id": "a4e227f506_0", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the comb appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a4e227f506_1", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. cable raceway\nB. eraser\nC. computer tower\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. cable raceway" - }, - { - "id": "a4e227f506_2", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the power socket, about 18 cm away. It is also to the side of and above the computer tower, about 69 cm away. It is further to the side of and above the eraser, about 78 cm away." - }, - { - "id": "a4e227f506_3", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"cable raceway\", \"references\": [{\"object\": \"power socket\", \"relation\": \"side_above\", \"distance_cm\": 17.9}, {\"object\": \"computer tower\", \"relation\": \"side_above\", \"distance_cm\": 69.0}]}" - }, - { - "id": "a4e227f506_4", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the notepad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a4e227f506_5", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "5", - "question_type": "target presence verification", - "question": "In the entire scene, did the eraser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "a4e227f506_6", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "6", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the eraser in this scene?\nA. cable raceway\nB. refrigerator\nC. power socket\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. cable raceway" - }, - { - "id": "a4e227f506_7", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the eraser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The eraser is to the side of and above the cable raceway, about 15 cm away. It is also to the side of and above the power socket, about 18 cm away. It is further to the side of and above the refrigerator, about 42 cm away." - }, - { - "id": "a4e227f506_8", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the eraser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"eraser\", \"support_surface\": \"power socket\", \"references\": [{\"object\": \"cable raceway\", \"relation\": \"side_above\", \"distance_cm\": 15.2}, {\"object\": \"power socket\", \"relation\": \"side_above\", \"distance_cm\": 18.4}]}" - }, - { - "id": "a4e227f506_9", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "9", - "question_type": "target presence verification", - "question": "In the entire scene, did the earphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a4e227f506_10", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "10", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. computer tower\nB. mouse\nC. monitor\nD. printer", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. monitor" - }, - { - "id": "a4e227f506_11", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "11", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is under the monitor, about 5 cm away. It is also next to the mouse, about 6 cm away. It is further to the side of and below the computer tower, about 23 cm away." - }, - { - "id": "a4e227f506_12", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "12", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"under\", \"distance_cm\": 5.0}, {\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 6.5}]}" - }, - { - "id": "a4e227f506_13", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "13", - "question_type": "target presence verification", - "question": "In the entire scene, did the tape dispenser appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "a4e227f506_14", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "14", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tape dispenser in this scene?\nA. chair\nB. monitor\nC. computer tower\nD. partition", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "a4e227f506_15", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "15", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tape dispenser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tape dispenser is to the side of and above the chair, about 18 cm away. It is also to the side of and above the computer tower, about 33 cm away. It is further to the side of and below the monitor, about 46 cm away." - }, - { - "id": "a4e227f506_16", - "dataset": "scannetpp", - "scene_name": "a4e227f506", - "sample_id": "16", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tape dispenser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tape dispenser\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 18.1}, {\"object\": \"computer tower\", \"relation\": \"side_above\", \"distance_cm\": 32.7}]}" - }, - { - "id": "a4f10da05d_0", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a4f10da05d_1", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. jacket\nB. light switches\nC. reusable shopping bag\nD. doorframe", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. reusable shopping bag" - }, - { - "id": "a4f10da05d_2", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is attached to the door. It is also to the side of and below the jacket, about 11 cm away. It is further near the doorframe, about 17 cm away." - }, - { - "id": "a4f10da05d_3", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"door\", \"references\": [{\"object\": \"jacket\", \"relation\": \"side_below\", \"distance_cm\": 10.8}, {\"object\": \"doorframe\", \"relation\": \"near\", \"distance_cm\": 16.8}]}" - }, - { - "id": "a4f10da05d_4", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the wireless charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a4f10da05d_5", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cable in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cable is to the side of and below the utility cart, about 8 cm away. It is also to the side of and below the cabinet, about 9 cm away. It is further to the side of and below the cosmetic bottle, about 47 cm away." - }, - { - "id": "a4f10da05d_6", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cable in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cable\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"utility cart\", \"relation\": \"side_below\", \"distance_cm\": 8.5}, {\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 8.6}]}" - }, - { - "id": "a4f10da05d_7", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a4f10da05d_8", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. monitor\nB. notebook\nC. mouse\nD. reusable cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. mouse" - }, - { - "id": "a4f10da05d_9", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also next to the mouse, about 4 cm away. It is further to the side of and below the reusable cup, about 6 cm away." - }, - { - "id": "a4f10da05d_10", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 4.4}, {\"object\": \"reusable cup\", \"relation\": \"side_below\", \"distance_cm\": 6.2}]}" - }, - { - "id": "a4f10da05d_11", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the power plug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a4f10da05d_12", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. keyboard\nB. clothes drying rack\nC. cloth\nD. reusable cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. keyboard" - }, - { - "id": "a4f10da05d_13", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also next to the keyboard, about 4 cm away. It is further to the side of and below the cloth, about 10 cm away." - }, - { - "id": "a4f10da05d_14", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 4.4}, {\"object\": \"cloth\", \"relation\": \"side_below\", \"distance_cm\": 9.7}]}" - }, - { - "id": "a4f10da05d_15", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the notebook appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "a4f10da05d_16", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the notebook in this scene?\nA. keyboard\nB. books\nC. reusable cup\nD. curtain", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. books" - }, - { - "id": "a4f10da05d_17", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the notebook in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The notebook is on the desk. It is also under the books, about 3 cm away. It is further to the side of and below the curtain, about 7 cm away." - }, - { - "id": "a4f10da05d_18", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the notebook in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"notebook\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"books\", \"relation\": \"under\", \"distance_cm\": 2.6}, {\"object\": \"curtain\", \"relation\": \"side_below\", \"distance_cm\": 7.5}]}" - }, - { - "id": "a4f10da05d_19", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "19", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "a4f10da05d_20", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "20", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pencil case in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pencil case is on the desk. It is also under the reusable cup, about 4 cm away. It is further under the table lamp, about 4 cm away." - }, - { - "id": "a4f10da05d_21", - "dataset": "scannetpp", - "scene_name": "a4f10da05d", - "sample_id": "21", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pencil case in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pencil case\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"reusable cup\", \"relation\": \"under\", \"distance_cm\": 3.9}, {\"object\": \"table lamp\", \"relation\": \"under\", \"distance_cm\": 3.9}]}" - }, - { - "id": "a7631adbe7_0", - "dataset": "scannetpp", - "scene_name": "a7631adbe7", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "a7631adbe7_1", - "dataset": "scannetpp", - "scene_name": "a7631adbe7", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. box\nB. chair\nC. coffee table\nD. coffee maker", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. chair" - }, - { - "id": "a7631adbe7_2", - "dataset": "scannetpp", - "scene_name": "a7631adbe7", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also to the side of and above the chair, about 25 cm away." - }, - { - "id": "a7631adbe7_3", - "dataset": "scannetpp", - "scene_name": "a7631adbe7", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 24.9}]}" - }, - { - "id": "a7631adbe7_4", - "dataset": "scannetpp", - "scene_name": "a7631adbe7", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the scissors appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a7631adbe7_5", - "dataset": "scannetpp", - "scene_name": "a7631adbe7", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. box\nB. washing machine\nC. coffee maker\nD. wet wipe", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. wet wipe" - }, - { - "id": "a7631adbe7_6", - "dataset": "scannetpp", - "scene_name": "a7631adbe7", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is on the coffee table. It is also next to the wet wipe, about 6 cm away. It is further near the box, about 20 cm away." - }, - { - "id": "a7631adbe7_7", - "dataset": "scannetpp", - "scene_name": "a7631adbe7", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"coffee table\", \"references\": [{\"object\": \"wet wipe\", \"relation\": \"next_to\", \"distance_cm\": 5.5}, {\"object\": \"box\", \"relation\": \"near\", \"distance_cm\": 20.4}]}" - }, - { - "id": "a766294249_0", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a766294249_1", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. laptop\nB. table\nC. curtain\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. table" - }, - { - "id": "a766294249_2", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is next to the table, about 5 cm away. It is also to the side of and below the laptop, about 11 cm away. It is further to the side of and above the chair, about 14 cm away." - }, - { - "id": "a766294249_3", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"next_to\", \"distance_cm\": 4.9}, {\"object\": \"laptop\", \"relation\": \"side_below\", \"distance_cm\": 10.8}]}" - }, - { - "id": "a766294249_4", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power plug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a766294249_5", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. book\nB. basket\nC. books\nD. spray bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. book" - }, - { - "id": "a766294249_6", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also to the side of and below the basket, about 6 cm away. It is further next to the books, about 7 cm away." - }, - { - "id": "a766294249_7", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"basket\", \"relation\": \"side_below\", \"distance_cm\": 5.9}, {\"object\": \"books\", \"relation\": \"next_to\", \"distance_cm\": 7.4}]}" - }, - { - "id": "a766294249_8", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the spray bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "a766294249_9", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the spray bottle in this scene?\nA. book\nB. laptop\nC. head phones\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. laptop" - }, - { - "id": "a766294249_10", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is next to the laptop, about 6 cm away. It is also to the side of and below the bottle, about 8 cm away. It is further to the side of and below the book, about 16 cm away." - }, - { - "id": "a766294249_11", - "dataset": "scannetpp", - "scene_name": "a766294249", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 6.3}, {\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 8.2}]}" - }, - { - "id": "a892730b61_0", - "dataset": "scannetpp", - "scene_name": "a892730b61", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "a892730b61_1", - "dataset": "scannetpp", - "scene_name": "a892730b61", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cable in this scene?\nA. wall lamp\nB. table\nC. keyboard\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. wall lamp" - }, - { - "id": "a892730b61_2", - "dataset": "scannetpp", - "scene_name": "a892730b61", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cable in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cable is to the side of and above the wall lamp, about 53 cm away." - }, - { - "id": "a892730b61_3", - "dataset": "scannetpp", - "scene_name": "a892730b61", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cable in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cable\", \"support_surface\": \"wall lamp\", \"references\": [{\"object\": \"wall lamp\", \"relation\": \"side_above\", \"distance_cm\": 53.0}]}" - }, - { - "id": "ab6983ae6c_0", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cable appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ab6983ae6c_1", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cable in this scene?\nA. headphone\nB. bottle\nC. monitor\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. table" - }, - { - "id": "ab6983ae6c_2", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cable in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cable is on the table. It is also near the headphone, about 26 cm away. It is further to the side of and above the office chair, about 42 cm away." - }, - { - "id": "ab6983ae6c_3", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cable in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cable\", \"support_surface\": \"table\", \"references\": [{\"object\": \"headphone\", \"relation\": \"near\", \"distance_cm\": 26.4}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 41.6}]}" - }, - { - "id": "ab6983ae6c_4", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the tote bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ab6983ae6c_5", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. bottle\nB. office chair\nC. monitor\nD. headphone", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. office chair" - }, - { - "id": "ab6983ae6c_6", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also to the side of and above the office chair, about 7 cm away. It is further to the side of and below the bottle, about 16 cm away." - }, - { - "id": "ab6983ae6c_7", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 7.0}, {\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 15.8}]}" - }, - { - "id": "ab6983ae6c_8", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the scissors appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ab6983ae6c_9", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. computer tower\nB. keyboard\nC. mouse\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. computer tower" - }, - { - "id": "ab6983ae6c_10", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also under the computer tower, about 3 cm away. It is further near the mouse, about 16 cm away." - }, - { - "id": "ab6983ae6c_11", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"computer tower\", \"relation\": \"under\", \"distance_cm\": 2.6}, {\"object\": \"mouse\", \"relation\": \"near\", \"distance_cm\": 15.6}]}" - }, - { - "id": "ab6983ae6c_12", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ab6983ae6c_13", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen holder in this scene?\nA. mug\nB. table\nC. outlet\nD. split", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. mug" - }, - { - "id": "ab6983ae6c_14", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen holder in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen holder is to the side of and above the mug, about 14 cm away. It is also to the side of and above the table, about 16 cm away. It is further to the side of and above the outlet, about 24 cm away." - }, - { - "id": "ab6983ae6c_15", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen holder in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen holder\", \"support_surface\": \"table\", \"references\": [{\"object\": \"mug\", \"relation\": \"side_above\", \"distance_cm\": 14.2}, {\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 15.6}]}" - }, - { - "id": "ab6983ae6c_16", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the usb hub appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ab6983ae6c_17", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "17", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue in this scene?\nA. paper\nB. box\nC. bottle\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. box" - }, - { - "id": "ab6983ae6c_18", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "18", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue is on the cabinet. It is also next to the box, about 4 cm away. It is further next to the paper, about 10 cm away." - }, - { - "id": "ab6983ae6c_19", - "dataset": "scannetpp", - "scene_name": "ab6983ae6c", - "sample_id": "19", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 3.5}, {\"object\": \"paper\", \"relation\": \"next_to\", \"distance_cm\": 9.7}]}" - }, - { - "id": "ad562ce09b_0", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mobile phone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ad562ce09b_1", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. headphones\nB. laptop\nC. power extender\nD. laptop charger", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. power extender" - }, - { - "id": "ad562ce09b_2", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the power extender, about 15 cm away. It is also to the side of and above the laptop charger, about 26 cm away. It is further to the side of and below the laptop, about 43 cm away." - }, - { - "id": "ad562ce09b_3", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"power extender\", \"references\": [{\"object\": \"power extender\", \"relation\": \"side_above\", \"distance_cm\": 15.2}, {\"object\": \"laptop charger\", \"relation\": \"side_above\", \"distance_cm\": 25.5}]}" - }, - { - "id": "ad562ce09b_4", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ad562ce09b_5", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. shelf\nB. vase\nC. books\nD. backpack", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. shelf" - }, - { - "id": "ad562ce09b_6", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the shelf. It is also to the side of and below the books, about 18 cm away. It is further to the side of and above the vase, about 54 cm away." - }, - { - "id": "ad562ce09b_7", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"books\", \"relation\": \"side_below\", \"distance_cm\": 17.5}, {\"object\": \"vase\", \"relation\": \"side_above\", \"distance_cm\": 53.6}]}" - }, - { - "id": "ad562ce09b_8", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ad562ce09b_9", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. laptop\nB. laptop charger\nC. mouse\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. mouse" - }, - { - "id": "ad562ce09b_10", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the table. It is also near the mouse, about 22 cm away. It is further to the side of and above the laptop charger, about 24 cm away." - }, - { - "id": "ad562ce09b_11", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"table\", \"references\": [{\"object\": \"mouse\", \"relation\": \"near\", \"distance_cm\": 22.2}, {\"object\": \"laptop charger\", \"relation\": \"side_above\", \"distance_cm\": 23.9}]}" - }, - { - "id": "ad562ce09b_12", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ad562ce09b_13", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. headphones\nB. laptop\nC. chair\nD. towel", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. laptop" - }, - { - "id": "ad562ce09b_14", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also next to the laptop, about 6 cm away. It is further near the headphones, about 22 cm away." - }, - { - "id": "ad562ce09b_15", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 5.5}, {\"object\": \"headphones\", \"relation\": \"near\", \"distance_cm\": 22.2}]}" - }, - { - "id": "ad562ce09b_16", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ad562ce09b_17", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone is on the bed. It is also next to the laptop, about 10 cm away. It is further next to the notebook, about 10 cm away." - }, - { - "id": "ad562ce09b_18", - "dataset": "scannetpp", - "scene_name": "ad562ce09b", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone\", \"support_surface\": \"bed\", \"references\": [{\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 9.6}, {\"object\": \"notebook\", \"relation\": \"next_to\", \"distance_cm\": 9.9}]}" - }, - { - "id": "af112b8903_0", - "dataset": "scannetpp", - "scene_name": "af112b8903", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "af112b8903_1", - "dataset": "scannetpp", - "scene_name": "af112b8903", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. cabinet\nB. cubicle separator\nC. table\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cubicle separator" - }, - { - "id": "af112b8903_2", - "dataset": "scannetpp", - "scene_name": "af112b8903", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also next to the cubicle separator, about 5 cm away. It is further to the side of and above the cabinet, about 13 cm away." - }, - { - "id": "af112b8903_3", - "dataset": "scannetpp", - "scene_name": "af112b8903", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"cubicle separator\", \"relation\": \"next_to\", \"distance_cm\": 4.8}, {\"object\": \"cabinet\", \"relation\": \"side_above\", \"distance_cm\": 13.1}]}" - }, - { - "id": "af112b8903_4", - "dataset": "scannetpp", - "scene_name": "af112b8903", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the toothbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "af112b8903_5", - "dataset": "scannetpp", - "scene_name": "af112b8903", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. cabinet\nB. rolling chair\nC. keyboard\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. keyboard" - }, - { - "id": "af112b8903_6", - "dataset": "scannetpp", - "scene_name": "af112b8903", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also near the keyboard, about 17 cm away. It is further to the side of and above the cabinet, about 22 cm away." - }, - { - "id": "af112b8903_7", - "dataset": "scannetpp", - "scene_name": "af112b8903", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"near\", \"distance_cm\": 17.2}, {\"object\": \"cabinet\", \"relation\": \"side_above\", \"distance_cm\": 21.9}]}" - }, - { - "id": "b0b004c40f_0", - "dataset": "scannetpp", - "scene_name": "b0b004c40f", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the post it note appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b0b004c40f_1", - "dataset": "scannetpp", - "scene_name": "b0b004c40f", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. chair\nB. table\nC. laptop\nD. duffel bag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. duffel bag" - }, - { - "id": "b0b004c40f_2", - "dataset": "scannetpp", - "scene_name": "b0b004c40f", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the table. It is also next to the duffel bag, about 11 cm away. It is further to the side of and above the laptop, about 13 cm away." - }, - { - "id": "b0b004c40f_3", - "dataset": "scannetpp", - "scene_name": "b0b004c40f", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"duffel bag\", \"relation\": \"next_to\", \"distance_cm\": 11.1}, {\"object\": \"laptop\", \"relation\": \"side_above\", \"distance_cm\": 12.6}]}" - }, - { - "id": "b2632b738a_0", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b2632b738a_1", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. water bottle\nB. tube\nC. dresser\nD. light bulb", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. dresser" - }, - { - "id": "b2632b738a_2", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the cabinet. It is also to the side of and above the dresser, about 8 cm away. It is further to the side of and below the water bottle, about 19 cm away." - }, - { - "id": "b2632b738a_3", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"dresser\", \"relation\": \"side_above\", \"distance_cm\": 8.4}, {\"object\": \"water bottle\", \"relation\": \"side_below\", \"distance_cm\": 18.6}]}" - }, - { - "id": "b2632b738a_4", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b2632b738a_5", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. desk\nB. plant pot\nC. desk organiser\nD. curtain", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. desk" - }, - { - "id": "b2632b738a_6", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the desk, about 4 cm away. It is also near the desk organiser, about 16 cm away. It is further to the side of and above the plant pot, about 19 cm away." - }, - { - "id": "b2632b738a_7", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"desk\", \"relation\": \"side_above\", \"distance_cm\": 4.3}, {\"object\": \"desk organiser\", \"relation\": \"near\", \"distance_cm\": 16.1}]}" - }, - { - "id": "b2632b738a_8", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "b2632b738a_9", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the charger in this scene?\nA. curtain\nB. table lamp\nC. desk organiser\nD. desk lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. table lamp" - }, - { - "id": "b2632b738a_10", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The charger is on the desk. It is also next to the table lamp, about 11 cm away. It is further to the side of and below the curtain, about 19 cm away." - }, - { - "id": "b2632b738a_11", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"charger\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"table lamp\", \"relation\": \"next_to\", \"distance_cm\": 10.7}, {\"object\": \"curtain\", \"relation\": \"side_below\", \"distance_cm\": 19.1}]}" - }, - { - "id": "b2632b738a_12", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b2632b738a_13", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. files\nB. bookshelf\nC. wheeled chair\nD. mirror", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. wheeled chair" - }, - { - "id": "b2632b738a_14", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is to the side of and above the wheeled chair, about 13 cm away. It is also to the side of and below the files, about 17 cm away. It is further to the side of and above the bookshelf, about 26 cm away." - }, - { - "id": "b2632b738a_15", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"wheeled chair\", \"relation\": \"side_above\", \"distance_cm\": 13.4}, {\"object\": \"files\", \"relation\": \"side_below\", \"distance_cm\": 16.6}]}" - }, - { - "id": "b2632b738a_16", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the scissors appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b2632b738a_17", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "17", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. tin box\nB. wall lamp\nC. umbrella\nD. shoes", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. tin box" - }, - { - "id": "b2632b738a_18", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "18", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the cabinet. It is also next to the tin box, about 3 cm away. It is further to the side of and above the umbrella, about 32 cm away." - }, - { - "id": "b2632b738a_19", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "19", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"tin box\", \"relation\": \"next_to\", \"distance_cm\": 2.8}, {\"object\": \"umbrella\", \"relation\": \"side_above\", \"distance_cm\": 31.8}]}" - }, - { - "id": "b2632b738a_20", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "20", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "b2632b738a_21", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "21", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the water bottle in this scene?\nA. tube\nB. backpack\nC. microwave\nD. light bulb", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. tube" - }, - { - "id": "b2632b738a_22", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "22", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water bottle is on the bookshelf. It is also to the side of and above the tube, about 7 cm away. It is further to the side of and above the microwave, about 17 cm away." - }, - { - "id": "b2632b738a_23", - "dataset": "scannetpp", - "scene_name": "b2632b738a", - "sample_id": "23", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water bottle\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"tube\", \"relation\": \"side_above\", \"distance_cm\": 6.8}, {\"object\": \"microwave\", \"relation\": \"side_above\", \"distance_cm\": 16.8}]}" - }, - { - "id": "b26e64c4b0_0", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "b26e64c4b0_1", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. box\nB. blinds\nC. shoe\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. shoe" - }, - { - "id": "b26e64c4b0_2", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the shoe, about 9 cm away. It is also to the side of and below the box, about 42 cm away." - }, - { - "id": "b26e64c4b0_3", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"box\", \"references\": [{\"object\": \"shoe\", \"relation\": \"side_above\", \"distance_cm\": 9.2}, {\"object\": \"box\", \"relation\": \"side_below\", \"distance_cm\": 42.0}]}" - }, - { - "id": "b26e64c4b0_4", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the lanyard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b26e64c4b0_5", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. guitar\nB. monitor\nC. blinds\nD. mouse", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. guitar" - }, - { - "id": "b26e64c4b0_6", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also to the side of and below the guitar, about 11 cm away. It is further next to the mouse, about 13 cm away." - }, - { - "id": "b26e64c4b0_7", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"guitar\", \"relation\": \"side_below\", \"distance_cm\": 10.7}, {\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 13.4}]}" - }, - { - "id": "b26e64c4b0_8", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "b26e64c4b0_9", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. plant\nB. mouse\nC. plant pot\nD. blinds", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. mouse" - }, - { - "id": "b26e64c4b0_10", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also next to the mouse, about 5 cm away. It is further to the side of and below the blinds, about 8 cm away." - }, - { - "id": "b26e64c4b0_11", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 5.0}, {\"object\": \"blinds\", \"relation\": \"side_below\", \"distance_cm\": 7.7}]}" - }, - { - "id": "b26e64c4b0_12", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b26e64c4b0_13", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is under the monitor, about 4 cm away. It is also next to the keyboard, about 5 cm away. It is further to the side of and above the table, about 6 cm away." - }, - { - "id": "b26e64c4b0_14", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"under\", \"distance_cm\": 4.0}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 5.0}]}" - }, - { - "id": "b26e64c4b0_15", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the eyeglasses appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b26e64c4b0_16", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse pad in this scene?\nA. chair\nB. pc\nC. guitar\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. monitor" - }, - { - "id": "b26e64c4b0_17", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse pad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse pad is on the table. It is also to the side of and below the monitor, about 6 cm away. It is further to the side of and below the guitar, about 10 cm away." - }, - { - "id": "b26e64c4b0_18", - "dataset": "scannetpp", - "scene_name": "b26e64c4b0", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse pad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse pad\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 6.1}, {\"object\": \"guitar\", \"relation\": \"side_below\", \"distance_cm\": 9.9}]}" - }, - { - "id": "b37c033757_0", - "dataset": "scannetpp", - "scene_name": "b37c033757", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the wireless charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b37c033757_1", - "dataset": "scannetpp", - "scene_name": "b37c033757", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. rolling cart\nB. cable\nC. vacuum cleaner\nD. barrel", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. barrel" - }, - { - "id": "b37c033757_2", - "dataset": "scannetpp", - "scene_name": "b37c033757", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is under the barrel, about 3 cm away. It is also to the side of and below the rolling cart, about 6 cm away. It is further to the side of and below the vacuum cleaner, about 23 cm away." - }, - { - "id": "b37c033757_3", - "dataset": "scannetpp", - "scene_name": "b37c033757", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"barrel\", \"references\": [{\"object\": \"barrel\", \"relation\": \"under\", \"distance_cm\": 3.1}, {\"object\": \"rolling cart\", \"relation\": \"side_below\", \"distance_cm\": 6.4}]}" - }, - { - "id": "b8d441d736_0", - "dataset": "scannetpp", - "scene_name": "b8d441d736", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the glasses appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "b8d441d736_1", - "dataset": "scannetpp", - "scene_name": "b8d441d736", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. headphone case\nB. clipboard\nC. duffel bag\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "b8d441d736_2", - "dataset": "scannetpp", - "scene_name": "b8d441d736", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the table. It is also to the side of and above the chair, about 19 cm away. It is further near the headphone case, about 27 cm away." - }, - { - "id": "b8d441d736_3", - "dataset": "scannetpp", - "scene_name": "b8d441d736", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 19.2}, {\"object\": \"headphone case\", \"relation\": \"near\", \"distance_cm\": 26.9}]}" - }, - { - "id": "b8d441d736_4", - "dataset": "scannetpp", - "scene_name": "b8d441d736", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphone case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "b8d441d736_5", - "dataset": "scannetpp", - "scene_name": "b8d441d736", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphone case in this scene?\nA. chair\nB. duffel bag\nC. backpack\nD. clipboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "b8d441d736_6", - "dataset": "scannetpp", - "scene_name": "b8d441d736", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphone case in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphone case is on the table. It is also to the side of and above the chair, about 16 cm away. It is further near the backpack, about 27 cm away." - }, - { - "id": "b8d441d736_7", - "dataset": "scannetpp", - "scene_name": "b8d441d736", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphone case in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphone case\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 16.1}, {\"object\": \"backpack\", \"relation\": \"near\", \"distance_cm\": 26.9}]}" - }, - { - "id": "bb05a0c48c_0", - "dataset": "scannetpp", - "scene_name": "bb05a0c48c", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "bb05a0c48c_1", - "dataset": "scannetpp", - "scene_name": "bb05a0c48c", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the box, about 21 cm away. It is also near the tube, about 23 cm away. It is further to the side of and above the file folder, about 26 cm away." - }, - { - "id": "bb05a0c48c_2", - "dataset": "scannetpp", - "scene_name": "bb05a0c48c", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"book shelf\", \"references\": [{\"object\": \"box\", \"relation\": \"side_above\", \"distance_cm\": 21.0}, {\"object\": \"tube\", \"relation\": \"near\", \"distance_cm\": 22.6}]}" - }, - { - "id": "bb05a0c48c_3", - "dataset": "scannetpp", - "scene_name": "bb05a0c48c", - "sample_id": "3", - "question_type": "target presence verification", - "question": "In the entire scene, did the lotion bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "bb05a0c48c_4", - "dataset": "scannetpp", - "scene_name": "bb05a0c48c", - "sample_id": "4", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. bedside table\nB. mop\nC. bed\nD. cleaning bucket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. bedside table" - }, - { - "id": "bb05a0c48c_5", - "dataset": "scannetpp", - "scene_name": "bb05a0c48c", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is under the bedside table. It is also to the side of and below the cleaning bucket, about 24 cm away. It is further to the side of and below the bed, about 32 cm away." - }, - { - "id": "bb05a0c48c_6", - "dataset": "scannetpp", - "scene_name": "bb05a0c48c", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"bedside table\", \"references\": [{\"object\": \"bedside table\", \"relation\": \"under\", \"distance_cm\": 2.3}, {\"object\": \"cleaning bucket\", \"relation\": \"side_below\", \"distance_cm\": 24.0}]}" - }, - { - "id": "bc400d86e1_0", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "bc400d86e1_1", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headset in this scene?\nA. computer tower\nB. cd player\nC. mouse\nD. mouse pad", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. mouse pad" - }, - { - "id": "bc400d86e1_2", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headset in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headset is next to the mouse pad, about 5 cm away. It is also to the side of and below the computer tower, about 6 cm away. It is further next to the cd player, about 7 cm away." - }, - { - "id": "bc400d86e1_3", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headset in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headset\", \"support_surface\": \"office table\", \"references\": [{\"object\": \"mouse pad\", \"relation\": \"next_to\", \"distance_cm\": 4.5}, {\"object\": \"computer tower\", \"relation\": \"side_below\", \"distance_cm\": 6.1}]}" - }, - { - "id": "bc400d86e1_4", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the wireless charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "bc400d86e1_5", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse pad in this scene?\nA. keyboard\nB. headset\nC. monitor\nD. cd player", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. headset" - }, - { - "id": "bc400d86e1_6", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse pad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse pad is next to the headset, about 5 cm away. It is also to the side of and below the cd player, about 6 cm away. It is further next to the keyboard, about 12 cm away." - }, - { - "id": "bc400d86e1_7", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse pad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse pad\", \"support_surface\": \"office table\", \"references\": [{\"object\": \"headset\", \"relation\": \"next_to\", \"distance_cm\": 4.5}, {\"object\": \"cd player\", \"relation\": \"side_below\", \"distance_cm\": 6.4}]}" - }, - { - "id": "bc400d86e1_8", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the stapler appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "bc400d86e1_9", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the stapler in this scene?\nA. box\nB. books\nC. storage cabinet\nD. folders", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. box" - }, - { - "id": "bc400d86e1_10", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the stapler in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The stapler is to the side of and above the box, about 20 cm away. It is also to the side of and below the books, about 26 cm away. It is further to the side of and above the storage cabinet, about 27 cm away." - }, - { - "id": "bc400d86e1_11", - "dataset": "scannetpp", - "scene_name": "bc400d86e1", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the stapler in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"stapler\", \"support_surface\": \"storage rack\", \"references\": [{\"object\": \"box\", \"relation\": \"side_above\", \"distance_cm\": 20.1}, {\"object\": \"books\", \"relation\": \"side_below\", \"distance_cm\": 26.4}]}" - }, - { - "id": "bcbc6e1d2d_0", - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "bcbc6e1d2d_1", - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. table\nB. chair\nC. notebook\nD. whiteboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "bcbc6e1d2d_2", - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the table, about 5 cm away. It is also to the side of and above the chair, about 37 cm away. It is further to the side of and below the whiteboard, about 91 cm away." - }, - { - "id": "bcbc6e1d2d_3", - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 4.9}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 37.0}]}" - }, - { - "id": "bcbc6e1d2d_4", - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphone case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "bcbc6e1d2d_5", - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the notebook in this scene?\nA. chair\nB. bottle\nC. table\nD. paper napkin", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "bcbc6e1d2d_6", - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the notebook in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The notebook is to the side of and above the chair, about 28 cm away." - }, - { - "id": "bcbc6e1d2d_7", - "dataset": "scannetpp", - "scene_name": "bcbc6e1d2d", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the notebook in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"notebook\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 27.9}]}" - }, - { - "id": "bd7375297e_0", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "bd7375297e_1", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pencil case in this scene?\nA. power adapter\nB. electrical outlet\nC. bottle\nD. paper", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. power adapter" - }, - { - "id": "bd7375297e_2", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pencil case in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pencil case is on the table. It is also next to the power adapter, about 8 cm away. It is further to the side of and below the bottle, about 13 cm away." - }, - { - "id": "bd7375297e_3", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pencil case in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pencil case\", \"support_surface\": \"table\", \"references\": [{\"object\": \"power adapter\", \"relation\": \"next_to\", \"distance_cm\": 8.2}, {\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 12.8}]}" - }, - { - "id": "bd7375297e_4", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the calculator appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "bd7375297e_5", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toothbrush in this scene?\nA. power socket\nB. soap dispenser\nC. object/toiletry\nD. sink", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. sink" - }, - { - "id": "bd7375297e_6", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toothbrush in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toothbrush is to the side of and above the sink, about 3 cm away. It is also under the soap dispenser, about 5 cm away. It is further next to the object/toiletry, about 7 cm away." - }, - { - "id": "bd7375297e_7", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toothbrush in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toothbrush\", \"support_surface\": \"bathroom cabinet/bathroom counter\", \"references\": [{\"object\": \"sink\", \"relation\": \"side_above\", \"distance_cm\": 3.2}, {\"object\": \"soap dispenser\", \"relation\": \"under\", \"distance_cm\": 4.8}]}" - }, - { - "id": "bd7375297e_8", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "bd7375297e_9", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tv remote in this scene?\nA. box\nB. tv board\nC. pencil case\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. box" - }, - { - "id": "bd7375297e_10", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tv remote in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tv remote is on the table. It is also to the side of and below the box, about 6 cm away. It is further to the side of and below the bottle, about 13 cm away." - }, - { - "id": "bd7375297e_11", - "dataset": "scannetpp", - "scene_name": "bd7375297e", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tv remote in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tv remote\", \"support_surface\": \"table\", \"references\": [{\"object\": \"box\", \"relation\": \"side_below\", \"distance_cm\": 5.9}, {\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 13.5}]}" - }, - { - "id": "bd9305480d_0", - "dataset": "scannetpp", - "scene_name": "bd9305480d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the purse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "bd9305480d_1", - "dataset": "scannetpp", - "scene_name": "bd9305480d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. paper\nB. locker\nC. bullettin boards\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. paper" - }, - { - "id": "bd9305480d_2", - "dataset": "scannetpp", - "scene_name": "bd9305480d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also to the side of and above the paper, about 6 cm away. It is further to the side of and above the locker, about 38 cm away." - }, - { - "id": "bd9305480d_3", - "dataset": "scannetpp", - "scene_name": "bd9305480d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"paper\", \"relation\": \"side_above\", \"distance_cm\": 6.3}, {\"object\": \"locker\", \"relation\": \"side_above\", \"distance_cm\": 38.3}]}" - }, - { - "id": "c06a983e63_0", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "c06a983e63_1", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. lamp\nB. monitor\nC. book\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. lamp" - }, - { - "id": "c06a983e63_2", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also under the lamp, about 5 cm away. It is further to the side of and below the monitor, about 22 cm away." - }, - { - "id": "c06a983e63_3", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"lamp\", \"relation\": \"under\", \"distance_cm\": 4.6}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 21.8}]}" - }, - { - "id": "c06a983e63_4", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "c06a983e63_5", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. folders\nB. office chair\nC. cabinet\nD. mouse", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. mouse" - }, - { - "id": "c06a983e63_6", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also next to the mouse, about 8 cm away. It is further near the folders, about 33 cm away." - }, - { - "id": "c06a983e63_7", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 7.7}, {\"object\": \"folders\", \"relation\": \"near\", \"distance_cm\": 32.8}]}" - }, - { - "id": "c06a983e63_8", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the adapter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c06a983e63_9", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. office chair\nB. cabinet\nC. keyboard\nD. phone", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. keyboard" - }, - { - "id": "c06a983e63_10", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also next to the keyboard, about 8 cm away. It is further to the side of and above the cabinet, about 27 cm away." - }, - { - "id": "c06a983e63_11", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 7.7}, {\"object\": \"cabinet\", \"relation\": \"side_above\", \"distance_cm\": 26.8}]}" - }, - { - "id": "c06a983e63_12", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the power bank appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c06a983e63_13", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone in this scene?\nA. book\nB. box\nC. lamp\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. lamp" - }, - { - "id": "c06a983e63_14", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone is on the table. It is also to the side of and below the lamp, about 6 cm away. It is further next to the box, about 15 cm away." - }, - { - "id": "c06a983e63_15", - "dataset": "scannetpp", - "scene_name": "c06a983e63", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone\", \"support_surface\": \"table\", \"references\": [{\"object\": \"lamp\", \"relation\": \"side_below\", \"distance_cm\": 6.2}, {\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 14.9}]}" - }, - { - "id": "c07c8a276d_0", - "dataset": "scannetpp", - "scene_name": "c07c8a276d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c07c8a276d_1", - "dataset": "scannetpp", - "scene_name": "c07c8a276d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. standing lamp\nB. cable strip\nC. power socket\nD. bed", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cable strip" - }, - { - "id": "c07c8a276d_2", - "dataset": "scannetpp", - "scene_name": "c07c8a276d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and above the cable strip, about 10 cm away. It is also to the side of and below the bed, about 30 cm away. It is further to the side of and below the standing lamp, about 37 cm away." - }, - { - "id": "c07c8a276d_3", - "dataset": "scannetpp", - "scene_name": "c07c8a276d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"bedside table\", \"references\": [{\"object\": \"cable strip\", \"relation\": \"side_above\", \"distance_cm\": 9.7}, {\"object\": \"bed\", \"relation\": \"side_below\", \"distance_cm\": 29.7}]}" - }, - { - "id": "c07c8a276d_4", - "dataset": "scannetpp", - "scene_name": "c07c8a276d", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "c07c8a276d_5", - "dataset": "scannetpp", - "scene_name": "c07c8a276d", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. blanket\nB. kitchen paper roll\nC. standing lamp\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. box" - }, - { - "id": "c07c8a276d_6", - "dataset": "scannetpp", - "scene_name": "c07c8a276d", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the desk. It is also next to the box, about 3 cm away. It is further next to the kitchen paper roll, about 6 cm away." - }, - { - "id": "c07c8a276d_7", - "dataset": "scannetpp", - "scene_name": "c07c8a276d", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 2.7}, {\"object\": \"kitchen paper roll\", \"relation\": \"next_to\", \"distance_cm\": 5.6}]}" - }, - { - "id": "c0c863b72d_0", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "c0c863b72d_1", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. paper\nB. chair\nC. cabinet\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. table" - }, - { - "id": "c0c863b72d_2", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the table, about 3 cm away. It is also to the side of and above the paper, about 35 cm away. It is further to the side of and above the cabinet, about 39 cm away." - }, - { - "id": "c0c863b72d_3", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 3.0}, {\"object\": \"paper\", \"relation\": \"side_above\", \"distance_cm\": 35.2}]}" - }, - { - "id": "c0c863b72d_4", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the purse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c0c863b72d_5", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse pad in this scene?\nA. cabinet\nB. office chair\nC. keyboard\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. keyboard" - }, - { - "id": "c0c863b72d_6", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse pad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse pad is on the table. It is also next to the keyboard, about 8 cm away. It is further to the side of and below the monitor, about 14 cm away." - }, - { - "id": "c0c863b72d_7", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse pad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse pad\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 8.3}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 13.6}]}" - }, - { - "id": "c0c863b72d_8", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c0c863b72d_9", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is to the side of and above the table, about 3 cm away. It is also to the side of and above the paper, about 3 cm away. It is further next to the box, about 4 cm away." - }, - { - "id": "c0c863b72d_10", - "dataset": "scannetpp", - "scene_name": "c0c863b72d", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 2.6}, {\"object\": \"paper\", \"relation\": \"side_above\", \"distance_cm\": 3.3}]}" - }, - { - "id": "c285c82ade_0", - "dataset": "scannetpp", - "scene_name": "c285c82ade", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the eyeglasses appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c285c82ade_1", - "dataset": "scannetpp", - "scene_name": "c285c82ade", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone charger in this scene?\nA. papers\nB. light switch\nC. doorframe\nD. window door", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. window door" - }, - { - "id": "c285c82ade_2", - "dataset": "scannetpp", - "scene_name": "c285c82ade", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is to the side of and below the door, about 10 cm away. It is also to the side of and below the window door, about 29 cm away. It is further below the light switch, about 73 cm away." - }, - { - "id": "c285c82ade_3", - "dataset": "scannetpp", - "scene_name": "c285c82ade", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"door\", \"references\": [{\"object\": \"door\", \"relation\": \"side_below\", \"distance_cm\": 9.9}, {\"object\": \"window door\", \"relation\": \"side_below\", \"distance_cm\": 29.1}]}" - }, - { - "id": "c285c82ade_4", - "dataset": "scannetpp", - "scene_name": "c285c82ade", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the water bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "c285c82ade_5", - "dataset": "scannetpp", - "scene_name": "c285c82ade", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the water bottle in this scene?\nA. doorframe\nB. window door\nC. papers\nD. light switch", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. doorframe" - }, - { - "id": "c285c82ade_6", - "dataset": "scannetpp", - "scene_name": "c285c82ade", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the water bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The water bottle is to the side of and below the doorframe, about 41 cm away. It is also to the side of and above the door, about 42 cm away. It is further to the side of and above the window door, about 65 cm away." - }, - { - "id": "c285c82ade_7", - "dataset": "scannetpp", - "scene_name": "c285c82ade", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the water bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"water bottle\", \"support_surface\": \"door\", \"references\": [{\"object\": \"doorframe\", \"relation\": \"side_below\", \"distance_cm\": 40.9}, {\"object\": \"door\", \"relation\": \"side_above\", \"distance_cm\": 41.8}]}" - }, - { - "id": "c2d714d386_0", - "dataset": "scannetpp", - "scene_name": "c2d714d386", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the comb appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c2d714d386_1", - "dataset": "scannetpp", - "scene_name": "c2d714d386", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the handbag in this scene?\nA. closet\nB. bookshelf\nC. table\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. bookshelf" - }, - { - "id": "c2d714d386_2", - "dataset": "scannetpp", - "scene_name": "c2d714d386", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the handbag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The handbag is to the side of and below the bookshelf, about 27 cm away." - }, - { - "id": "c2d714d386_3", - "dataset": "scannetpp", - "scene_name": "c2d714d386", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the handbag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"handbag\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"bookshelf\", \"relation\": \"side_below\", \"distance_cm\": 27.3}]}" - }, - { - "id": "c413b34238_0", - "dataset": "scannetpp", - "scene_name": "c413b34238", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "c413b34238_1", - "dataset": "scannetpp", - "scene_name": "c413b34238", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. decoration\nB. plant pot\nC. whiteboard\nD. plant", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. plant" - }, - { - "id": "c413b34238_2", - "dataset": "scannetpp", - "scene_name": "c413b34238", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the cabinet. It is also under the plant, about 4 cm away. It is further next to the plant pot, about 7 cm away." - }, - { - "id": "c413b34238_3", - "dataset": "scannetpp", - "scene_name": "c413b34238", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"plant\", \"relation\": \"under\", \"distance_cm\": 4.0}, {\"object\": \"plant pot\", \"relation\": \"next_to\", \"distance_cm\": 7.3}]}" - }, - { - "id": "c413b34238_4", - "dataset": "scannetpp", - "scene_name": "c413b34238", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power adapter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c413b34238_5", - "dataset": "scannetpp", - "scene_name": "c413b34238", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. keyboard\nB. chair\nC. ring light\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. keyboard" - }, - { - "id": "c413b34238_6", - "dataset": "scannetpp", - "scene_name": "c413b34238", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is next to the keyboard, about 5 cm away. It is also to the side of and below the monitor, about 7 cm away. It is further to the side of and above the chair, about 18 cm away." - }, - { - "id": "c413b34238_7", - "dataset": "scannetpp", - "scene_name": "c413b34238", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 5.3}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 7.5}]}" - }, - { - "id": "c465f388d1_0", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the extension cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c465f388d1_1", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. beverage bottle\nB. mattress\nC. bedside table\nD. pillow", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. beverage bottle" - }, - { - "id": "c465f388d1_2", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is on the carpet. It is also to the side of and above the beverage bottle, about 3 cm away. It is further to the side of and below the mattress, about 6 cm away." - }, - { - "id": "c465f388d1_3", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"carpet\", \"references\": [{\"object\": \"beverage bottle\", \"relation\": \"side_above\", \"distance_cm\": 3.4}, {\"object\": \"mattress\", \"relation\": \"side_below\", \"distance_cm\": 6.4}]}" - }, - { - "id": "c465f388d1_4", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "c465f388d1_5", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. laptop\nB. box\nC. mouse\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. laptop" - }, - { - "id": "c465f388d1_6", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also next to the laptop, about 7 cm away. It is further next to the box, about 13 cm away." - }, - { - "id": "c465f388d1_7", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 7.0}, {\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 12.6}]}" - }, - { - "id": "c465f388d1_8", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphone case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c465f388d1_9", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone charger in this scene?\nA. air freshener\nB. paper bag\nC. book\nD. pillow", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. air freshener" - }, - { - "id": "c465f388d1_10", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is to the side of and above the air freshener, about 12 cm away. It is also to the side of and below the paper bag, about 29 cm away. It is further near the pillow, about 35 cm away." - }, - { - "id": "c465f388d1_11", - "dataset": "scannetpp", - "scene_name": "c465f388d1", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"pillow\", \"references\": [{\"object\": \"air freshener\", \"relation\": \"side_above\", \"distance_cm\": 12.2}, {\"object\": \"paper bag\", \"relation\": \"side_below\", \"distance_cm\": 28.8}]}" - }, - { - "id": "c856c41c99_0", - "dataset": "scannetpp", - "scene_name": "c856c41c99", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the stapler appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c856c41c99_1", - "dataset": "scannetpp", - "scene_name": "c856c41c99", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the spray bottle in this scene?\nA. board\nB. dumbbell case\nC. monitor\nD. dumbbell", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. dumbbell" - }, - { - "id": "c856c41c99_2", - "dataset": "scannetpp", - "scene_name": "c856c41c99", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is next to the dumbbell, about 10 cm away. It is also to the side of and below the dumbbell case, about 56 cm away." - }, - { - "id": "c856c41c99_3", - "dataset": "scannetpp", - "scene_name": "c856c41c99", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"dumbbell case\", \"references\": [{\"object\": \"dumbbell\", \"relation\": \"next_to\", \"distance_cm\": 9.9}, {\"object\": \"dumbbell case\", \"relation\": \"side_below\", \"distance_cm\": 56.4}]}" - }, - { - "id": "c8d099ecd8_0", - "dataset": "scannetpp", - "scene_name": "c8d099ecd8", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the wallet appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "c8d099ecd8_1", - "dataset": "scannetpp", - "scene_name": "c8d099ecd8", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the extension cord in this scene?\nA. office chair\nB. desk\nC. extension board\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. office chair" - }, - { - "id": "c8d099ecd8_2", - "dataset": "scannetpp", - "scene_name": "c8d099ecd8", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the extension cord in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The extension cord is on the table. It is also to the side of and above the office chair, about 12 cm away." - }, - { - "id": "c8d099ecd8_3", - "dataset": "scannetpp", - "scene_name": "c8d099ecd8", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the extension cord in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"extension cord\", \"support_surface\": \"table\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 12.3}]}" - }, - { - "id": "c8d099ecd8_4", - "dataset": "scannetpp", - "scene_name": "c8d099ecd8", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "c8d099ecd8_5", - "dataset": "scannetpp", - "scene_name": "c8d099ecd8", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen in this scene?\nA. tea cup\nB. telephone\nC. papers\nD. picture", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. papers" - }, - { - "id": "c8d099ecd8_6", - "dataset": "scannetpp", - "scene_name": "c8d099ecd8", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen is on the desk. It is also next to the papers, about 7 cm away. It is further next to the telephone, about 10 cm away." - }, - { - "id": "c8d099ecd8_7", - "dataset": "scannetpp", - "scene_name": "c8d099ecd8", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"papers\", \"relation\": \"next_to\", \"distance_cm\": 6.7}, {\"object\": \"telephone\", \"relation\": \"next_to\", \"distance_cm\": 9.8}]}" - }, - { - "id": "ca0c580422_0", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ca0c580422_1", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. post it notes\nB. monitor\nC. office chair\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. monitor" - }, - { - "id": "ca0c580422_2", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the table. It is also to the side of and below the monitor, about 17 cm away. It is further near the cup, about 20 cm away." - }, - { - "id": "ca0c580422_3", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 16.6}, {\"object\": \"cup\", \"relation\": \"near\", \"distance_cm\": 19.5}]}" - }, - { - "id": "ca0c580422_4", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ca0c580422_5", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. post it notes\nB. box\nC. bottle\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. post it notes" - }, - { - "id": "ca0c580422_6", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also next to the post it notes, about 15 cm away. It is further near the bottle, about 20 cm away." - }, - { - "id": "ca0c580422_7", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"post it notes\", \"relation\": \"next_to\", \"distance_cm\": 14.6}, {\"object\": \"bottle\", \"relation\": \"near\", \"distance_cm\": 19.5}]}" - }, - { - "id": "ca0c580422_8", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse pad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ca0c580422_9", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is to the side of and above the table, about 4 cm away. It is also to the side of and above the cable, about 5 cm away. It is further to the side of and below the window sill, about 9 cm away." - }, - { - "id": "ca0c580422_10", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 4.1}, {\"object\": \"cable\", \"relation\": \"side_above\", \"distance_cm\": 4.6}]}" - }, - { - "id": "ca0c580422_11", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the stapler appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ca0c580422_12", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the stapler in this scene?\nA. monitor\nB. post it notes\nC. table\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. post it notes" - }, - { - "id": "ca0c580422_13", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the stapler in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The stapler is near the post it notes, about 17 cm away. It is also to the side of and above the table, about 22 cm away. It is further to the side of and below the monitor, about 22 cm away." - }, - { - "id": "ca0c580422_14", - "dataset": "scannetpp", - "scene_name": "ca0c580422", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the stapler in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"stapler\", \"support_surface\": \"table\", \"references\": [{\"object\": \"post it notes\", \"relation\": \"near\", \"distance_cm\": 16.9}, {\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 21.9}]}" - }, - { - "id": "ce12db9e81_0", - "dataset": "scannetpp", - "scene_name": "ce12db9e81", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the adapter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ce12db9e81_1", - "dataset": "scannetpp", - "scene_name": "ce12db9e81", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the adapter in this scene?\nA. heater\nB. light switches\nC. chair\nD. heat pipe", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. light switches" - }, - { - "id": "ce12db9e81_2", - "dataset": "scannetpp", - "scene_name": "ce12db9e81", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the adapter in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The adapter is near the light switches, about 18 cm away. It is also to the side of and above the chair, about 50 cm away." - }, - { - "id": "ce12db9e81_3", - "dataset": "scannetpp", - "scene_name": "ce12db9e81", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the adapter in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"adapter\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"light switches\", \"relation\": \"near\", \"distance_cm\": 18.3}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 49.7}]}" - }, - { - "id": "ce12db9e81_4", - "dataset": "scannetpp", - "scene_name": "ce12db9e81", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse pad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ce12db9e81_5", - "dataset": "scannetpp", - "scene_name": "ce12db9e81", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power adapter in this scene?\nA. power sockets\nB. heater\nC. heat pipe\nD. heater pipe", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. heater" - }, - { - "id": "ce12db9e81_6", - "dataset": "scannetpp", - "scene_name": "ce12db9e81", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power adapter in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power adapter is to the side of and above the heater, about 20 cm away. It is also to the side of and above the heat pipe, about 51 cm away. It is further to the side of and above the heater pipe, about 61 cm away." - }, - { - "id": "ce12db9e81_7", - "dataset": "scannetpp", - "scene_name": "ce12db9e81", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power adapter in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power adapter\", \"support_surface\": \"power sockets\", \"references\": [{\"object\": \"heater\", \"relation\": \"side_above\", \"distance_cm\": 19.7}, {\"object\": \"heat pipe\", \"relation\": \"side_above\", \"distance_cm\": 51.2}]}" - }, - { - "id": "d6d9ddb03f_0", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "d6d9ddb03f_1", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "1", - "question_type": "target presence verification", - "question": "In the entire scene, did the charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "d6d9ddb03f_2", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "2", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. office chair\nB. box\nC. backpack\nD. packet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. box" - }, - { - "id": "d6d9ddb03f_3", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "3", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the box, about 16 cm away. It is also to the side of and above the office chair, about 45 cm away. It is further near the packet, about 48 cm away." - }, - { - "id": "d6d9ddb03f_4", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "4", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"box\", \"relation\": \"side_above\", \"distance_cm\": 15.6}, {\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 44.6}]}" - }, - { - "id": "d6d9ddb03f_5", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "5", - "question_type": "target presence verification", - "question": "In the entire scene, did the smartphone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "d6d9ddb03f_6", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "6", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. keyboard\nB. file\nC. monitor\nD. mouse", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. mouse" - }, - { - "id": "d6d9ddb03f_7", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also next to the mouse, about 13 cm away. It is further near the keyboard, about 15 cm away." - }, - { - "id": "d6d9ddb03f_8", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 13.4}, {\"object\": \"keyboard\", \"relation\": \"near\", \"distance_cm\": 15.3}]}" - }, - { - "id": "d6d9ddb03f_9", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "9", - "question_type": "target presence verification", - "question": "In the entire scene, did the headset appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "d6d9ddb03f_10", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "10", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headset in this scene?\nA. power socket\nB. keyboard\nC. monitor\nD. table lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. power socket" - }, - { - "id": "d6d9ddb03f_11", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "11", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headset in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headset is near the power socket, about 16 cm away. It is also to the side of and below the table lamp, about 17 cm away. It is further to the side of and below the monitor, about 20 cm away." - }, - { - "id": "d6d9ddb03f_12", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "12", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headset in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headset\", \"support_surface\": \"table lamp\", \"references\": [{\"object\": \"power socket\", \"relation\": \"near\", \"distance_cm\": 15.5}, {\"object\": \"table lamp\", \"relation\": \"side_below\", \"distance_cm\": 16.9}]}" - }, - { - "id": "d6d9ddb03f_13", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "13", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "d6d9ddb03f_14", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "14", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. mouse\nB. cup\nC. office chair\nD. file", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. mouse" - }, - { - "id": "d6d9ddb03f_15", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "15", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also next to the mouse, about 6 cm away. It is further near the cup, about 15 cm away." - }, - { - "id": "d6d9ddb03f_16", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "16", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 5.7}, {\"object\": \"cup\", \"relation\": \"near\", \"distance_cm\": 15.3}]}" - }, - { - "id": "d6d9ddb03f_17", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "17", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "d6d9ddb03f_18", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "18", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. keyboard\nB. office chair\nC. file\nD. cup", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. keyboard" - }, - { - "id": "d6d9ddb03f_19", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "19", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also next to the keyboard, about 6 cm away. It is further next to the cup, about 13 cm away." - }, - { - "id": "d6d9ddb03f_20", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "20", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 5.7}, {\"object\": \"cup\", \"relation\": \"next_to\", \"distance_cm\": 13.4}]}" - }, - { - "id": "d6d9ddb03f_21", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "21", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "d6d9ddb03f_22", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "22", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the notebook in this scene?\nA. cup\nB. cabinet\nC. file\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. box" - }, - { - "id": "d6d9ddb03f_23", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "23", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the notebook in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The notebook is on the table. It is also to the side of and below the box, about 7 cm away. It is further to the side of and above the cabinet, about 24 cm away." - }, - { - "id": "d6d9ddb03f_24", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "24", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the notebook in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"notebook\", \"support_surface\": \"table\", \"references\": [{\"object\": \"box\", \"relation\": \"side_below\", \"distance_cm\": 7.0}, {\"object\": \"cabinet\", \"relation\": \"side_above\", \"distance_cm\": 24.3}]}" - }, - { - "id": "d6d9ddb03f_25", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "25", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "d6d9ddb03f_26", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "26", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. heater\nB. table\nC. box\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. heater" - }, - { - "id": "d6d9ddb03f_27", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "27", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the heater, about 41 cm away. It is also to the side of and below the box, about 47 cm away. It is further to the side of and below the table, about 51 cm away." - }, - { - "id": "d6d9ddb03f_28", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "28", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"heater\", \"relation\": \"side_below\", \"distance_cm\": 41.4}, {\"object\": \"box\", \"relation\": \"side_below\", \"distance_cm\": 46.7}]}" - }, - { - "id": "d6d9ddb03f_29", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "29", - "question_type": "target presence verification", - "question": "In the entire scene, did the handbag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "d6d9ddb03f_30", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "30", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the remote control in this scene?\nA. jacket\nB. umbrella\nC. suitcase\nD. storage cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. storage cabinet" - }, - { - "id": "d6d9ddb03f_31", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "31", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the remote control in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The remote control is on the cabinet. It is also to the side of and below the storage cabinet, about 5 cm away. It is further to the side of and below the jacket, about 15 cm away." - }, - { - "id": "d6d9ddb03f_32", - "dataset": "scannetpp", - "scene_name": "d6d9ddb03f", - "sample_id": "32", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the remote control in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"remote control\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"storage cabinet\", \"relation\": \"side_below\", \"distance_cm\": 5.1}, {\"object\": \"jacket\", \"relation\": \"side_below\", \"distance_cm\": 14.9}]}" - }, - { - "id": "d918af9c5f_0", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "d918af9c5f_1", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. monitor\nB. keyboard\nC. wheel\nD. picture frame", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. picture frame" - }, - { - "id": "d918af9c5f_2", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the desk. It is also next to the picture frame, about 4 cm away. It is further to the side of and below the monitor, about 12 cm away." - }, - { - "id": "d918af9c5f_3", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"picture frame\", \"relation\": \"next_to\", \"distance_cm\": 4.1}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 12.5}]}" - }, - { - "id": "d918af9c5f_4", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power bank appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "d918af9c5f_5", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the desk. It is also to the side of and below the monitor, about 8 cm away. It is further to the side of and below the laptop, about 8 cm away." - }, - { - "id": "d918af9c5f_6", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 8.2}, {\"object\": \"laptop\", \"relation\": \"side_below\", \"distance_cm\": 8.3}]}" - }, - { - "id": "d918af9c5f_7", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "d918af9c5f_8", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. keyboard\nB. chair\nC. laptop\nD. desk lamp", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. laptop" - }, - { - "id": "d918af9c5f_9", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also to the side of and below the laptop, about 11 cm away. It is further next to the keyboard, about 14 cm away." - }, - { - "id": "d918af9c5f_10", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"laptop\", \"relation\": \"side_below\", \"distance_cm\": 10.8}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 13.6}]}" - }, - { - "id": "d918af9c5f_11", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "11", - "question_type": "target presence verification", - "question": "In the entire scene, did the toothbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "d918af9c5f_12", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "12", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the toothbrush in this scene?\nA. bathroom cabinet\nB. tooth paste\nC. power sockets\nD. shelf", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. tooth paste" - }, - { - "id": "d918af9c5f_13", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the toothbrush in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The toothbrush is next to the tooth paste. It is also to the side of and below the shelf, about 12 cm away. It is further to the side of and below the power sockets, about 13 cm away." - }, - { - "id": "d918af9c5f_14", - "dataset": "scannetpp", - "scene_name": "d918af9c5f", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the toothbrush in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"toothbrush\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"tooth paste\", \"relation\": \"next_to\", \"distance_cm\": 2.1}, {\"object\": \"shelf\", \"relation\": \"side_below\", \"distance_cm\": 11.7}]}" - }, - { - "id": "dbd97b5bca_0", - "dataset": "scannetpp", - "scene_name": "dbd97b5bca", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the glue bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "dbd97b5bca_1", - "dataset": "scannetpp", - "scene_name": "dbd97b5bca", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. storage rack\nB. toilet paper\nC. bottle\nD. shelf rail", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. shelf rail" - }, - { - "id": "dbd97b5bca_2", - "dataset": "scannetpp", - "scene_name": "dbd97b5bca", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is under the shelf rail, about 3 cm away. It is also to the side of and below the bottle, about 11 cm away. It is further next to the storage rack, about 11 cm away." - }, - { - "id": "dbd97b5bca_3", - "dataset": "scannetpp", - "scene_name": "dbd97b5bca", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"shelf rail\", \"references\": [{\"object\": \"shelf rail\", \"relation\": \"under\", \"distance_cm\": 3.1}, {\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 10.8}]}" - }, - { - "id": "dbd97b5bca_4", - "dataset": "scannetpp", - "scene_name": "dbd97b5bca", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "dbd97b5bca_5", - "dataset": "scannetpp", - "scene_name": "dbd97b5bca", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. heater\nB. towel\nC. bedsheet\nD. window frame", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. bedsheet" - }, - { - "id": "dbd97b5bca_6", - "dataset": "scannetpp", - "scene_name": "dbd97b5bca", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is on the table. It is also to the side of and above the bedsheet, about 14 cm away. It is further to the side of and above the heater, about 25 cm away." - }, - { - "id": "dbd97b5bca_7", - "dataset": "scannetpp", - "scene_name": "dbd97b5bca", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"bedsheet\", \"relation\": \"side_above\", \"distance_cm\": 13.8}, {\"object\": \"heater\", \"relation\": \"side_above\", \"distance_cm\": 25.4}]}" - }, - { - "id": "dd636ce603_0", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the deodorant bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "dd636ce603_1", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. intercom\nB. cleaning supply\nC. power socket\nD. doorframe", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cleaning supply" - }, - { - "id": "dd636ce603_2", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the cleaning supply, about 12 cm away. It is also to the side of and below the intercom, about 20 cm away. It is further to the side of and below the doorframe, about 25 cm away." - }, - { - "id": "dd636ce603_3", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"cleaning supply\", \"relation\": \"side_above\", \"distance_cm\": 12.1}, {\"object\": \"intercom\", \"relation\": \"side_below\", \"distance_cm\": 19.5}]}" - }, - { - "id": "dd636ce603_4", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the glue bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "dd636ce603_5", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. monitor\nB. jar\nC. milk carton\nD. fruits", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. jar" - }, - { - "id": "dd636ce603_6", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is next to the jar. It is also to the side of and below the monitor, about 6 cm away. It is further next to the milk carton, about 7 cm away." - }, - { - "id": "dd636ce603_7", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"jar\", \"relation\": \"next_to\", \"distance_cm\": 2.5}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 6.1}]}" - }, - { - "id": "dd636ce603_8", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the notebook appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "dd636ce603_9", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the notebook in this scene?\nA. mouse\nB. milk carton\nC. mug\nD. tissue", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. mouse" - }, - { - "id": "dd636ce603_10", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the notebook in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The notebook is next to the mouse, about 8 cm away. It is also to the side of and below the mug, about 11 cm away. It is further to the side of and below the tissue, about 16 cm away." - }, - { - "id": "dd636ce603_11", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the notebook in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"notebook\", \"support_surface\": \"bookshelf\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 8.3}, {\"object\": \"mug\", \"relation\": \"side_below\", \"distance_cm\": 11.2}]}" - }, - { - "id": "dd636ce603_12", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the spray bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "dd636ce603_13", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the spray bottle in this scene?\nA. power adapter\nB. power strip\nC. table\nD. laptop", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. power strip" - }, - { - "id": "dd636ce603_14", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the spray bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The spray bottle is next to the window, about 3 cm away. It is also to the side of and above the power strip, about 13 cm away. It is further to the side of and above the table, about 29 cm away." - }, - { - "id": "dd636ce603_15", - "dataset": "scannetpp", - "scene_name": "dd636ce603", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the spray bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"spray bottle\", \"support_surface\": \"table\", \"references\": [{\"object\": \"window\", \"relation\": \"next_to\", \"distance_cm\": 3.2}, {\"object\": \"power strip\", \"relation\": \"side_above\", \"distance_cm\": 13.2}]}" - }, - { - "id": "de5881aa12_0", - "dataset": "scannetpp", - "scene_name": "de5881aa12", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "de5881aa12_1", - "dataset": "scannetpp", - "scene_name": "de5881aa12", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. bedside table\nB. speaker/projector\nC. pillow/cushion\nD. sofa", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. sofa" - }, - { - "id": "de5881aa12_2", - "dataset": "scannetpp", - "scene_name": "de5881aa12", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is under the sofa, about 3 cm away. It is also to the side of and below the pillow/cushion, about 27 cm away. It is further to the side of and below the speaker/projector, about 32 cm away." - }, - { - "id": "de5881aa12_3", - "dataset": "scannetpp", - "scene_name": "de5881aa12", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"bedside table\", \"references\": [{\"object\": \"sofa\", \"relation\": \"under\", \"distance_cm\": 2.8}, {\"object\": \"pillow/cushion\", \"relation\": \"side_below\", \"distance_cm\": 27.5}]}" - }, - { - "id": "de5881aa12_4", - "dataset": "scannetpp", - "scene_name": "de5881aa12", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the calculator appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "de5881aa12_5", - "dataset": "scannetpp", - "scene_name": "de5881aa12", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the earphones in this scene?\nA. box\nB. chair\nC. stationery\nD. laptop", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. laptop" - }, - { - "id": "de5881aa12_6", - "dataset": "scannetpp", - "scene_name": "de5881aa12", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the earphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The earphones is next to the laptop, about 5 cm away. It is also to the side of and above the chair, about 26 cm away. It is further to the side of and below the stationery, about 34 cm away." - }, - { - "id": "de5881aa12_7", - "dataset": "scannetpp", - "scene_name": "de5881aa12", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the earphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"earphones\", \"support_surface\": \"table\", \"references\": [{\"object\": \"laptop\", \"relation\": \"next_to\", \"distance_cm\": 5.1}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 26.0}]}" - }, - { - "id": "dfac5b38df_0", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the earbuds appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "dfac5b38df_1", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. cabinet\nB. office chair\nC. food pot\nD. coaster", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. coaster" - }, - { - "id": "dfac5b38df_2", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and above the coaster, about 6 cm away. It is also to the side of and above the food pot, about 15 cm away. It is further to the side of and above the office chair, about 36 cm away." - }, - { - "id": "dfac5b38df_3", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"coaster\", \"relation\": \"side_above\", \"distance_cm\": 6.2}, {\"object\": \"food pot\", \"relation\": \"side_above\", \"distance_cm\": 15.4}]}" - }, - { - "id": "dfac5b38df_4", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the scissors appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "dfac5b38df_5", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. power adapter\nB. monitor\nC. keyboard\nD. mug", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. monitor" - }, - { - "id": "dfac5b38df_6", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is to the side of and below the monitor, about 7 cm away. It is also next to the mug, about 10 cm away. It is further next to the power adapter, about 12 cm away." - }, - { - "id": "dfac5b38df_7", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 7.1}, {\"object\": \"mug\", \"relation\": \"next_to\", \"distance_cm\": 10.3}]}" - }, - { - "id": "dfac5b38df_8", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the usb hub appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "dfac5b38df_9", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. mouse\nB. office chair\nC. monitor\nD. headphones", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. mouse" - }, - { - "id": "dfac5b38df_10", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is next to the mouse, about 8 cm away. It is also to the side of and below the monitor, about 9 cm away. It is further next to the headphones, about 14 cm away." - }, - { - "id": "dfac5b38df_11", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 8.1}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 9.2}]}" - }, - { - "id": "dfac5b38df_12", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "dfac5b38df_13", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "13", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is next to the keyboard, about 8 cm away. It is also to the side of and below the monitor, about 8 cm away. It is further to the side of and above the office chair, about 31 cm away." - }, - { - "id": "dfac5b38df_14", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "14", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"office chair\", \"references\": [{\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 8.1}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 8.1}]}" - }, - { - "id": "dfac5b38df_15", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "15", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "dfac5b38df_16", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "16", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. office chair\nB. cabinet\nC. headphones\nD. power adapter", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. headphones" - }, - { - "id": "dfac5b38df_17", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "17", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is next to the headphones, about 10 cm away. It is also to the side of and above the cabinet, about 17 cm away. It is further to the side of and above the office chair, about 21 cm away." - }, - { - "id": "dfac5b38df_18", - "dataset": "scannetpp", - "scene_name": "dfac5b38df", - "sample_id": "18", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"headphones\", \"relation\": \"next_to\", \"distance_cm\": 10.3}, {\"object\": \"cabinet\", \"relation\": \"side_above\", \"distance_cm\": 16.7}]}" - }, - { - "id": "e0de253456_0", - "dataset": "scannetpp", - "scene_name": "e0de253456", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the toothbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e0de253456_1", - "dataset": "scannetpp", - "scene_name": "e0de253456", - "sample_id": "1", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the soap in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The soap is under the tile wall, about 3 cm away. It is also next to the split, about 3 cm away. It is further under the cream tube, about 3 cm away." - }, - { - "id": "e0de253456_2", - "dataset": "scannetpp", - "scene_name": "e0de253456", - "sample_id": "2", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the soap in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"soap\", \"support_surface\": \"tile wall\", \"references\": [{\"object\": \"tile wall\", \"relation\": \"under\", \"distance_cm\": 2.5}, {\"object\": \"split\", \"relation\": \"next_to\", \"distance_cm\": 2.7}]}" - }, - { - "id": "e3b3b0d0c7_0", - "dataset": "scannetpp", - "scene_name": "e3b3b0d0c7", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the wallet appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e3b3b0d0c7_1", - "dataset": "scannetpp", - "scene_name": "e3b3b0d0c7", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. table\nB. keyboard\nC. doorframe\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. doorframe" - }, - { - "id": "e3b3b0d0c7_2", - "dataset": "scannetpp", - "scene_name": "e3b3b0d0c7", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is to the side of and above the doorframe, about 16 cm away. It is also near the door, about 22 cm away." - }, - { - "id": "e3b3b0d0c7_3", - "dataset": "scannetpp", - "scene_name": "e3b3b0d0c7", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"door\", \"references\": [{\"object\": \"doorframe\", \"relation\": \"side_above\", \"distance_cm\": 16.3}, {\"object\": \"door\", \"relation\": \"near\", \"distance_cm\": 21.6}]}" - }, - { - "id": "e3ef8b690b_0", - "dataset": "scannetpp", - "scene_name": "e3ef8b690b", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the highlighter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e3ef8b690b_1", - "dataset": "scannetpp", - "scene_name": "e3ef8b690b", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. bouquet\nB. chair\nC. plate\nD. tablecloth", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. bouquet" - }, - { - "id": "e3ef8b690b_2", - "dataset": "scannetpp", - "scene_name": "e3ef8b690b", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also under the bouquet, about 5 cm away. It is further to the side of and above the chair, about 31 cm away." - }, - { - "id": "e3ef8b690b_3", - "dataset": "scannetpp", - "scene_name": "e3ef8b690b", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"bouquet\", \"relation\": \"under\", \"distance_cm\": 4.9}, {\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 31.0}]}" - }, - { - "id": "e3ef8b690b_4", - "dataset": "scannetpp", - "scene_name": "e3ef8b690b", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e3ef8b690b_5", - "dataset": "scannetpp", - "scene_name": "e3ef8b690b", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tv remote in this scene?\nA. sofa cushion\nB. bouquet\nC. blanket\nD. pillow", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. pillow" - }, - { - "id": "e3ef8b690b_6", - "dataset": "scannetpp", - "scene_name": "e3ef8b690b", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tv remote in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tv remote is to the side of and below the pillow, about 6 cm away. It is also to the side of and above the blanket, about 15 cm away. It is further to the side of and below the sofa cushion, about 26 cm away." - }, - { - "id": "e3ef8b690b_7", - "dataset": "scannetpp", - "scene_name": "e3ef8b690b", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tv remote in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tv remote\", \"support_surface\": \"sofa\", \"references\": [{\"object\": \"pillow\", \"relation\": \"side_below\", \"distance_cm\": 5.8}, {\"object\": \"blanket\", \"relation\": \"side_above\", \"distance_cm\": 15.3}]}" - }, - { - "id": "e51b753e48_0", - "dataset": "scannetpp", - "scene_name": "e51b753e48", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the key appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e51b753e48_1", - "dataset": "scannetpp", - "scene_name": "e51b753e48", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. book\nB. plant\nC. lounge chair\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. book" - }, - { - "id": "e51b753e48_2", - "dataset": "scannetpp", - "scene_name": "e51b753e48", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is next to the book. It is also next to the plant, about 5 cm away. It is further to the side of and above the lounge chair, about 9 cm away." - }, - { - "id": "e51b753e48_3", - "dataset": "scannetpp", - "scene_name": "e51b753e48", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"bedside table\", \"references\": [{\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 2.2}, {\"object\": \"plant\", \"relation\": \"next_to\", \"distance_cm\": 5.0}]}" - }, - { - "id": "e51b753e48_4", - "dataset": "scannetpp", - "scene_name": "e51b753e48", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e51b753e48_5", - "dataset": "scannetpp", - "scene_name": "e51b753e48", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue in this scene?\nA. bed\nB. poster\nC. windowsill\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. windowsill" - }, - { - "id": "e51b753e48_6", - "dataset": "scannetpp", - "scene_name": "e51b753e48", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue is to the side of and below the windowsill, about 10 cm away. It is also to the side of and below the poster, about 15 cm away. It is further to the side of and above the bed, about 17 cm away." - }, - { - "id": "e51b753e48_7", - "dataset": "scannetpp", - "scene_name": "e51b753e48", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue\", \"support_surface\": \"bedside table\", \"references\": [{\"object\": \"windowsill\", \"relation\": \"side_below\", \"distance_cm\": 10.0}, {\"object\": \"poster\", \"relation\": \"side_below\", \"distance_cm\": 15.0}]}" - }, - { - "id": "e6372539da_0", - "dataset": "scannetpp", - "scene_name": "e6372539da", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e6372539da_1", - "dataset": "scannetpp", - "scene_name": "e6372539da", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. cabinet\nB. bag\nC. washing machine\nD. light switches", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. washing machine" - }, - { - "id": "e6372539da_2", - "dataset": "scannetpp", - "scene_name": "e6372539da", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is to the side of and below the washing machine, about 35 cm away. It is also to the side of and below the light switches, about 73 cm away." - }, - { - "id": "e6372539da_3", - "dataset": "scannetpp", - "scene_name": "e6372539da", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"washing machine\", \"relation\": \"side_below\", \"distance_cm\": 34.9}, {\"object\": \"light switches\", \"relation\": \"side_below\", \"distance_cm\": 72.7}]}" - }, - { - "id": "e6372539da_4", - "dataset": "scannetpp", - "scene_name": "e6372539da", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e6372539da_5", - "dataset": "scannetpp", - "scene_name": "e6372539da", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. towel\nB. bag\nC. cloth\nD. drying rack", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. drying rack" - }, - { - "id": "e6372539da_6", - "dataset": "scannetpp", - "scene_name": "e6372539da", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is next to the drying rack, about 12 cm away. It is also to the side of and above the bag, about 33 cm away. It is further to the side of and below the towel, about 51 cm away." - }, - { - "id": "e6372539da_7", - "dataset": "scannetpp", - "scene_name": "e6372539da", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"drying rack\", \"references\": [{\"object\": \"drying rack\", \"relation\": \"next_to\", \"distance_cm\": 11.7}, {\"object\": \"bag\", \"relation\": \"side_above\", \"distance_cm\": 33.2}]}" - }, - { - "id": "e667e09fe6_0", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pen holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e667e09fe6_1", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. bedside cabinet\nB. mug\nC. wardrobe\nD. shampoo bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. bedside cabinet" - }, - { - "id": "e667e09fe6_2", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is to the side of and above the bedside cabinet, about 5 cm away. It is also to the side of and below the wardrobe, about 10 cm away. It is further to the side of and above the mug, about 12 cm away." - }, - { - "id": "e667e09fe6_3", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"bedside cabinet\", \"references\": [{\"object\": \"bedside cabinet\", \"relation\": \"side_above\", \"distance_cm\": 5.5}, {\"object\": \"wardrobe\", \"relation\": \"side_below\", \"distance_cm\": 10.0}]}" - }, - { - "id": "e667e09fe6_4", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the keyboard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e667e09fe6_5", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. usb hub\nB. laptop\nC. mousepad\nD. mouse", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. mousepad" - }, - { - "id": "e667e09fe6_6", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is next to the mousepad, about 4 cm away. It is also next to the usb hub, about 8 cm away. It is further next to the mouse, about 9 cm away." - }, - { - "id": "e667e09fe6_7", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"mousepad\", \"relation\": \"next_to\", \"distance_cm\": 3.6}, {\"object\": \"usb hub\", \"relation\": \"next_to\", \"distance_cm\": 7.8}]}" - }, - { - "id": "e667e09fe6_8", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e667e09fe6_9", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the laptop charger in this scene?\nA. plastic bin\nB. trash can\nC. coffe table\nD. electric kettle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. coffe table" - }, - { - "id": "e667e09fe6_10", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the laptop charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The laptop charger is to the side of and below the coffe table, about 6 cm away. It is also to the side of and below the trash can, about 23 cm away. It is further to the side of and below the electric kettle, about 36 cm away." - }, - { - "id": "e667e09fe6_11", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the laptop charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"laptop charger\", \"support_surface\": \"coffe table\", \"references\": [{\"object\": \"coffe table\", \"relation\": \"side_below\", \"distance_cm\": 5.8}, {\"object\": \"trash can\", \"relation\": \"side_below\", \"distance_cm\": 22.9}]}" - }, - { - "id": "e667e09fe6_12", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "12", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e667e09fe6_13", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "13", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. laptop\nB. office chair\nC. keyboard\nD. glass", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. office chair" - }, - { - "id": "e667e09fe6_14", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "14", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the desk. It is also to the side of and above the office chair, about 7 cm away. It is further next to the keyboard, about 9 cm away." - }, - { - "id": "e667e09fe6_15", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "15", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 7.0}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 8.6}]}" - }, - { - "id": "e667e09fe6_16", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "16", - "question_type": "target presence verification", - "question": "In the entire scene, did the mousepad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e667e09fe6_17", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "17", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mousepad in this scene?\nA. laptop\nB. keyboard\nC. office chair\nD. glass", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. glass" - }, - { - "id": "e667e09fe6_18", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "18", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mousepad in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mousepad is under the glass. It is also next to the keyboard, about 4 cm away. It is further to the side of and below the laptop, about 6 cm away." - }, - { - "id": "e667e09fe6_19", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "19", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mousepad in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mousepad\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"glass\", \"relation\": \"under\", \"distance_cm\": 2.3}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 3.6}]}" - }, - { - "id": "e667e09fe6_20", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "20", - "question_type": "target presence verification", - "question": "In the entire scene, did the toothbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e667e09fe6_21", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "21", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. poster\nB. bedside cabinet\nC. headphones\nD. shampoo bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. headphones" - }, - { - "id": "e667e09fe6_22", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "22", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is to the side of and below the headphones, about 12 cm away. It is also below the poster, about 25 cm away. It is further to the side of and below the shampoo bottle, about 26 cm away." - }, - { - "id": "e667e09fe6_23", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "23", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"bedside cabinet\", \"references\": [{\"object\": \"headphones\", \"relation\": \"side_below\", \"distance_cm\": 11.5}, {\"object\": \"poster\", \"relation\": \"below\", \"distance_cm\": 25.4}]}" - }, - { - "id": "e667e09fe6_24", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "24", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote control appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e667e09fe6_25", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "25", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone charger in this scene?\nA. office chair\nB. desk\nC. reusable shopping bag\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. chair" - }, - { - "id": "e667e09fe6_26", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "26", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is to the side of and below the chair, about 17 cm away. It is also near the rug, about 20 cm away. It is further to the side of and below the desk, about 29 cm away." - }, - { - "id": "e667e09fe6_27", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "27", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 16.5}, {\"object\": \"rug\", \"relation\": \"near\", \"distance_cm\": 19.5}]}" - }, - { - "id": "e667e09fe6_28", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "28", - "question_type": "target presence verification", - "question": "In the entire scene, did the tote bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e667e09fe6_29", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "29", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the usb hub in this scene?\nA. laptop\nB. monitor\nC. controller\nD. keyboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. controller" - }, - { - "id": "e667e09fe6_30", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "30", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the usb hub in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The usb hub is next to the controller. It is also next to the keyboard, about 8 cm away. It is further to the side of and below the monitor, about 14 cm away." - }, - { - "id": "e667e09fe6_31", - "dataset": "scannetpp", - "scene_name": "e667e09fe6", - "sample_id": "31", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the usb hub in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"usb hub\", \"support_surface\": \"desk\", \"references\": [{\"object\": \"controller\", \"relation\": \"next_to\", \"distance_cm\": 2.5}, {\"object\": \"keyboard\", \"relation\": \"next_to\", \"distance_cm\": 7.8}]}" - }, - { - "id": "e7d7e7799d_0", - "dataset": "scannetpp", - "scene_name": "e7d7e7799d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the hairbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e7d7e7799d_1", - "dataset": "scannetpp", - "scene_name": "e7d7e7799d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tv remote in this scene?\nA. table\nB. blanket\nC. table lamp\nD. pillow", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "e7d7e7799d_2", - "dataset": "scannetpp", - "scene_name": "e7d7e7799d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tv remote in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tv remote is to the side of and above the table, about 12 cm away. It is also to the side of and above the blanket, about 17 cm away. It is further to the side of and below the pillow, about 17 cm away." - }, - { - "id": "e7d7e7799d_3", - "dataset": "scannetpp", - "scene_name": "e7d7e7799d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tv remote in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tv remote\", \"support_surface\": \"bedside cabinet\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 12.2}, {\"object\": \"blanket\", \"relation\": \"side_above\", \"distance_cm\": 16.7}]}" - }, - { - "id": "e8ea9b4da8_0", - "dataset": "scannetpp", - "scene_name": "e8ea9b4da8", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the key appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e8ea9b4da8_1", - "dataset": "scannetpp", - "scene_name": "e8ea9b4da8", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the wallet in this scene?\nA. light switch\nB. bowl\nC. decoration\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. book" - }, - { - "id": "e8ea9b4da8_2", - "dataset": "scannetpp", - "scene_name": "e8ea9b4da8", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the wallet in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The wallet is next to the book, about 3 cm away. It is also near the decoration, about 17 cm away. It is further to the side of and above the light switch, about 31 cm away." - }, - { - "id": "e8ea9b4da8_3", - "dataset": "scannetpp", - "scene_name": "e8ea9b4da8", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the wallet in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"wallet\", \"support_surface\": \"storage cabinet\", \"references\": [{\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 2.6}, {\"object\": \"decoration\", \"relation\": \"near\", \"distance_cm\": 17.4}]}" - }, - { - "id": "e909f8213d_0", - "dataset": "scannetpp", - "scene_name": "e909f8213d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the comb appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e909f8213d_1", - "dataset": "scannetpp", - "scene_name": "e909f8213d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the bottle in this scene?\nA. bag\nB. heater\nC. cutting board\nD. doorframe", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. cutting board" - }, - { - "id": "e909f8213d_2", - "dataset": "scannetpp", - "scene_name": "e909f8213d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the bottle in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The bottle is on the tray. It is also to the side of and above the cutting board, about 3 cm away. It is further to the side of and below the doorframe, about 14 cm away." - }, - { - "id": "e909f8213d_3", - "dataset": "scannetpp", - "scene_name": "e909f8213d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the bottle in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"bottle\", \"support_surface\": \"tray\", \"references\": [{\"object\": \"cutting board\", \"relation\": \"side_above\", \"distance_cm\": 3.0}, {\"object\": \"doorframe\", \"relation\": \"side_below\", \"distance_cm\": 13.7}]}" - }, - { - "id": "e909f8213d_4", - "dataset": "scannetpp", - "scene_name": "e909f8213d", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e909f8213d_5", - "dataset": "scannetpp", - "scene_name": "e909f8213d", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. sink\nB. sponge\nC. faucet\nD. toaster", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. sponge" - }, - { - "id": "e909f8213d_6", - "dataset": "scannetpp", - "scene_name": "e909f8213d", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is to the side of and above the sponge, about 8 cm away. It is also next to the faucet, about 12 cm away. It is further to the side of and above the sink, about 14 cm away." - }, - { - "id": "e909f8213d_7", - "dataset": "scannetpp", - "scene_name": "e909f8213d", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"kitchen counter\", \"references\": [{\"object\": \"sponge\", \"relation\": \"side_above\", \"distance_cm\": 8.1}, {\"object\": \"faucet\", \"relation\": \"next_to\", \"distance_cm\": 12.0}]}" - }, - { - "id": "e91722b5a3_0", - "dataset": "scannetpp", - "scene_name": "e91722b5a3", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the mousepad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e91722b5a3_1", - "dataset": "scannetpp", - "scene_name": "e91722b5a3", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the stapler in this scene?\nA. envelope\nB. monitor\nC. paper\nD. keyboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. monitor" - }, - { - "id": "e91722b5a3_2", - "dataset": "scannetpp", - "scene_name": "e91722b5a3", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the stapler in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The stapler is on the table. It is also to the side of and below the monitor, about 9 cm away. It is further next to the paper, about 13 cm away." - }, - { - "id": "e91722b5a3_3", - "dataset": "scannetpp", - "scene_name": "e91722b5a3", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the stapler in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"stapler\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 9.0}, {\"object\": \"paper\", \"relation\": \"next_to\", \"distance_cm\": 13.4}]}" - }, - { - "id": "e93cd5230e_0", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e93cd5230e_1", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "1", - "question_type": "target presence verification", - "question": "In the entire scene, did the laptop charger appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e93cd5230e_2", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "2", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the laptop charger in this scene?\nA. backpack\nB. heater\nC. power strip\nD. bag", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. power strip" - }, - { - "id": "e93cd5230e_3", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "3", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the laptop charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The laptop charger is next to the power strip, about 5 cm away. It is also near the bag, about 46 cm away. It is further to the side of and below the heater, about 50 cm away." - }, - { - "id": "e93cd5230e_4", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "4", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the laptop charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"laptop charger\", \"support_surface\": \"bag\", \"references\": [{\"object\": \"power strip\", \"relation\": \"next_to\", \"distance_cm\": 5.2}, {\"object\": \"bag\", \"relation\": \"near\", \"distance_cm\": 46.1}]}" - }, - { - "id": "e93cd5230e_5", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "5", - "question_type": "target presence verification", - "question": "In the entire scene, did the power plug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e93cd5230e_6", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "6", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. table\nB. heater\nC. windowsill\nD. laptop", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. table" - }, - { - "id": "e93cd5230e_7", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is to the side of and above the table, about 4 cm away. It is also near the laptop, about 15 cm away. It is further to the side of and above the heater, about 31 cm away." - }, - { - "id": "e93cd5230e_8", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_above\", \"distance_cm\": 4.2}, {\"object\": \"laptop\", \"relation\": \"near\", \"distance_cm\": 15.0}]}" - }, - { - "id": "e93cd5230e_9", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "9", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "e93cd5230e_10", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "10", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. power sockets\nB. router\nC. coax outlet\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. bottle" - }, - { - "id": "e93cd5230e_11", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "11", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also to the side of and below the window, about 12 cm away. It is further to the side of and below the bottle, about 29 cm away." - }, - { - "id": "e93cd5230e_12", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "12", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"window\", \"relation\": \"side_below\", \"distance_cm\": 11.7}, {\"object\": \"bottle\", \"relation\": \"side_below\", \"distance_cm\": 29.2}]}" - }, - { - "id": "e93cd5230e_13", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "13", - "question_type": "target presence verification", - "question": "In the entire scene, did the toothbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "e93cd5230e_14", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "14", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. laptop charger\nB. backpack\nC. bag\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. laptop charger" - }, - { - "id": "e93cd5230e_15", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "15", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is next to the laptop charger, about 5 cm away. It is also near the bag, about 44 cm away. It is further to the side of and below the backpack, about 47 cm away." - }, - { - "id": "e93cd5230e_16", - "dataset": "scannetpp", - "scene_name": "e93cd5230e", - "sample_id": "16", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"bag\", \"references\": [{\"object\": \"laptop charger\", \"relation\": \"next_to\", \"distance_cm\": 5.2}, {\"object\": \"bag\", \"relation\": \"near\", \"distance_cm\": 44.0}]}" - }, - { - "id": "ed2216380b_0", - "dataset": "scannetpp", - "scene_name": "ed2216380b", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the backpack appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ed2216380b_1", - "dataset": "scannetpp", - "scene_name": "ed2216380b", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. box\nB. bottle\nC. chair\nD. envelope", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. chair" - }, - { - "id": "ed2216380b_2", - "dataset": "scannetpp", - "scene_name": "ed2216380b", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the envelope, about 51 cm away." - }, - { - "id": "ed2216380b_3", - "dataset": "scannetpp", - "scene_name": "ed2216380b", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"table\", \"references\": [{\"object\": \"envelope\", \"relation\": \"side_below\", \"distance_cm\": 51.5}]}" - }, - { - "id": "ed2216380b_4", - "dataset": "scannetpp", - "scene_name": "ed2216380b", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the remote controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ed2216380b_5", - "dataset": "scannetpp", - "scene_name": "ed2216380b", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the pen holder in this scene?\nA. box\nB. whiteboard eraser\nC. backpack\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. whiteboard eraser" - }, - { - "id": "ed2216380b_6", - "dataset": "scannetpp", - "scene_name": "ed2216380b", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen holder in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen holder is next to the whiteboard eraser." - }, - { - "id": "ed2216380b_7", - "dataset": "scannetpp", - "scene_name": "ed2216380b", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen holder in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen holder\", \"support_surface\": \"whiteboard eraser\", \"references\": [{\"object\": \"whiteboard eraser\", \"relation\": \"next_to\", \"distance_cm\": 2.4}]}" - }, - { - "id": "eea4ad9c04_0", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the extension cord appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "eea4ad9c04_1", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the extension cord in this scene?\nA. chair\nB. cabinet\nC. lantern\nD. vase", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cabinet" - }, - { - "id": "eea4ad9c04_2", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the extension cord in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The extension cord is to the side of and below the cabinet, about 22 cm away. It is also near the vase, about 32 cm away. It is further to the side of and below the chair, about 42 cm away." - }, - { - "id": "eea4ad9c04_3", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the extension cord in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"extension cord\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 22.1}, {\"object\": \"vase\", \"relation\": \"near\", \"distance_cm\": 32.1}]}" - }, - { - "id": "eea4ad9c04_4", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "eea4ad9c04_5", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the laptop charger in this scene?\nA. books\nB. phone charger\nC. vase\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. phone charger" - }, - { - "id": "eea4ad9c04_6", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the laptop charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The laptop charger is to the side of and below the cabinet, about 11 cm away. It is also to the side of and below the vase, about 21 cm away. It is further to the side of and below the books, about 56 cm away." - }, - { - "id": "eea4ad9c04_7", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the laptop charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"laptop charger\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 11.1}, {\"object\": \"vase\", \"relation\": \"side_below\", \"distance_cm\": 21.4}]}" - }, - { - "id": "eea4ad9c04_8", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "eea4ad9c04_9", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is next to the trolley, about 6 cm away. It is also to the side of and below the digital piano cover, about 7 cm away. It is further near the bedside table, about 24 cm away." - }, - { - "id": "eea4ad9c04_10", - "dataset": "scannetpp", - "scene_name": "eea4ad9c04", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"bedside table\", \"references\": [{\"object\": \"trolley\", \"relation\": \"next_to\", \"distance_cm\": 6.1}, {\"object\": \"digital piano cover\", \"relation\": \"side_below\", \"distance_cm\": 7.1}]}" - }, - { - "id": "eee7eef8e5_0", - "dataset": "scannetpp", - "scene_name": "eee7eef8e5", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the calculator appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "eee7eef8e5_1", - "dataset": "scannetpp", - "scene_name": "eee7eef8e5", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the remote in this scene?\nA. chair\nB. table\nC. monitor\nD. split", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "eee7eef8e5_2", - "dataset": "scannetpp", - "scene_name": "eee7eef8e5", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the remote in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The remote is to the side of and above the chair, about 19 cm away." - }, - { - "id": "eee7eef8e5_3", - "dataset": "scannetpp", - "scene_name": "eee7eef8e5", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the remote in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"remote\", \"support_surface\": \"chair\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 19.2}]}" - }, - { - "id": "ef18cf0708_0", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ef18cf0708_1", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "1", - "question_type": "target presence verification", - "question": "In the entire scene, did the cable appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "ef18cf0708_2", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "2", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cable in this scene?\nA. electrical plug\nB. monitor\nC. power socket\nD. mouse", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. monitor" - }, - { - "id": "ef18cf0708_3", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "3", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cable in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cable is on the table. It is also under the monitor, about 3 cm away. It is further to the side of and below the electrical plug, about 19 cm away." - }, - { - "id": "ef18cf0708_4", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "4", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cable in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cable\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"under\", \"distance_cm\": 2.9}, {\"object\": \"electrical plug\", \"relation\": \"side_below\", \"distance_cm\": 19.2}]}" - }, - { - "id": "ef18cf0708_5", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "5", - "question_type": "target presence verification", - "question": "In the entire scene, did the comb appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ef18cf0708_6", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "6", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. power socket\nB. windowsill\nC. window sill\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. heater" - }, - { - "id": "ef18cf0708_7", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "7", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is to the side of and above the heater, about 28 cm away. It is also to the side of and below the power socket, about 36 cm away. It is further to the side of and below the window sill, about 42 cm away." - }, - { - "id": "ef18cf0708_8", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "8", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"heater\", \"relation\": \"side_above\", \"distance_cm\": 27.8}, {\"object\": \"power socket\", \"relation\": \"side_below\", \"distance_cm\": 36.5}]}" - }, - { - "id": "ef18cf0708_9", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "9", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse pad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "ef18cf0708_10", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "10", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mug in this scene?\nA. windowsill\nB. split\nC. telephone\nD. power socket", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. windowsill" - }, - { - "id": "ef18cf0708_11", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "11", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is to the side of and above the windowsill, about 3 cm away. It is also next to the telephone, about 6 cm away. It is further to the side of and above the power socket, about 21 cm away." - }, - { - "id": "ef18cf0708_12", - "dataset": "scannetpp", - "scene_name": "ef18cf0708", - "sample_id": "12", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"shelf\", \"references\": [{\"object\": \"windowsill\", \"relation\": \"side_above\", \"distance_cm\": 3.0}, {\"object\": \"telephone\", \"relation\": \"next_to\", \"distance_cm\": 6.4}]}" - }, - { - "id": "f19ca0a52e_0", - "dataset": "scannetpp", - "scene_name": "f19ca0a52e", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv remote appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "f19ca0a52e_1", - "dataset": "scannetpp", - "scene_name": "f19ca0a52e", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. power outlet\nB. cutting board\nC. kettle\nD. dishwasher", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. kettle" - }, - { - "id": "f19ca0a52e_2", - "dataset": "scannetpp", - "scene_name": "f19ca0a52e", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the cabinet. It is also to the side of and below the kettle, about 6 cm away. It is further to the side of and above the dishwasher, about 32 cm away." - }, - { - "id": "f19ca0a52e_3", - "dataset": "scannetpp", - "scene_name": "f19ca0a52e", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"kettle\", \"relation\": \"side_below\", \"distance_cm\": 6.2}, {\"object\": \"dishwasher\", \"relation\": \"side_above\", \"distance_cm\": 32.2}]}" - }, - { - "id": "f3d64c30f8_0", - "dataset": "scannetpp", - "scene_name": "f3d64c30f8", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "f3d64c30f8_1", - "dataset": "scannetpp", - "scene_name": "f3d64c30f8", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. jug\nB. box\nC. cabinet\nD. whiteboard", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. box" - }, - { - "id": "f3d64c30f8_2", - "dataset": "scannetpp", - "scene_name": "f3d64c30f8", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is next to the box, about 7 cm away. It is also to the side of and above the jug, about 24 cm away. It is further to the side of and below the whiteboard, about 25 cm away." - }, - { - "id": "f3d64c30f8_3", - "dataset": "scannetpp", - "scene_name": "f3d64c30f8", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 7.2}, {\"object\": \"jug\", \"relation\": \"side_above\", \"distance_cm\": 24.2}]}" - }, - { - "id": "f3d64c30f8_4", - "dataset": "scannetpp", - "scene_name": "f3d64c30f8", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "f3d64c30f8_5", - "dataset": "scannetpp", - "scene_name": "f3d64c30f8", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. bookshelf\nB. power socket\nC. bag\nD. bottle", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. power socket" - }, - { - "id": "f3d64c30f8_6", - "dataset": "scannetpp", - "scene_name": "f3d64c30f8", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the power socket, about 50 cm away." - }, - { - "id": "f3d64c30f8_7", - "dataset": "scannetpp", - "scene_name": "f3d64c30f8", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"power socket\", \"references\": [{\"object\": \"power socket\", \"relation\": \"side_below\", \"distance_cm\": 50.3}]}" - }, - { - "id": "f3f016ba3f_0", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the headphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "f3f016ba3f_1", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the headphones in this scene?\nA. pen holder\nB. monitor\nC. toy\nD. office chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. office chair" - }, - { - "id": "f3f016ba3f_2", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the headphones in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The headphones is on the table. It is also to the side of and above the office chair, about 3 cm away. It is further to the side of and below the monitor, about 12 cm away." - }, - { - "id": "f3f016ba3f_3", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the headphones in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"headphones\", \"support_surface\": \"table\", \"references\": [{\"object\": \"office chair\", \"relation\": \"side_above\", \"distance_cm\": 3.0}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 11.7}]}" - }, - { - "id": "f3f016ba3f_4", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil holder appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "f3f016ba3f_5", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the pen holder in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The pen holder is on the table. It is also to the side of and below the monitor, about 10 cm away. It is further to the side of and above the toy, about 10 cm away." - }, - { - "id": "f3f016ba3f_6", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the pen holder in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"pen holder\", \"support_surface\": \"table\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 9.6}, {\"object\": \"toy\", \"relation\": \"side_above\", \"distance_cm\": 10.1}]}" - }, - { - "id": "f3f016ba3f_7", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "f3f016ba3f_8", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. picture\nB. yoga mat\nC. toy\nD. table", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. table" - }, - { - "id": "f3f016ba3f_9", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the table, about 46 cm away. It is also near the yoga mat, about 48 cm away. It is further to the side of and below the toy, about 67 cm away." - }, - { - "id": "f3f016ba3f_10", - "dataset": "scannetpp", - "scene_name": "f3f016ba3f", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 46.2}, {\"object\": \"yoga mat\", \"relation\": \"near\", \"distance_cm\": 47.5}]}" - }, - { - "id": "f6659a3107_0", - "dataset": "scannetpp", - "scene_name": "f6659a3107", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the glue bottle appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "f6659a3107_1", - "dataset": "scannetpp", - "scene_name": "f6659a3107", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the eraser in this scene?\nA. table\nB. keyboard\nC. split\nD. monitor", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. split" - }, - { - "id": "f6659a3107_2", - "dataset": "scannetpp", - "scene_name": "f6659a3107", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the eraser in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The eraser is to the side of and above the split, about 26 cm away." - }, - { - "id": "f6659a3107_3", - "dataset": "scannetpp", - "scene_name": "f6659a3107", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the eraser in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"eraser\", \"support_surface\": \"split\", \"references\": [{\"object\": \"split\", \"relation\": \"side_above\", \"distance_cm\": 26.0}]}" - }, - { - "id": "f6a9b64a0d_0", - "dataset": "scannetpp", - "scene_name": "f6a9b64a0d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power adapter appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "f6a9b64a0d_1", - "dataset": "scannetpp", - "scene_name": "f6a9b64a0d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. bedside lamp\nB. table\nC. plug\nD. book", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. table" - }, - { - "id": "f6a9b64a0d_2", - "dataset": "scannetpp", - "scene_name": "f6a9b64a0d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the table, about 6 cm away. It is also to the side of and below the book, about 30 cm away. It is further to the side of and below the bedside lamp, about 34 cm away." - }, - { - "id": "f6a9b64a0d_3", - "dataset": "scannetpp", - "scene_name": "f6a9b64a0d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"table\", \"references\": [{\"object\": \"table\", \"relation\": \"side_below\", \"distance_cm\": 5.7}, {\"object\": \"book\", \"relation\": \"side_below\", \"distance_cm\": 30.2}]}" - }, - { - "id": "f8d5147d1d_0", - "dataset": "scannetpp", - "scene_name": "f8d5147d1d", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the power strip appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "f8d5147d1d_1", - "dataset": "scannetpp", - "scene_name": "f8d5147d1d", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the power strip in this scene?\nA. monitor\nB. stapler\nC. office table\nD. hole puncher", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. office table" - }, - { - "id": "f8d5147d1d_2", - "dataset": "scannetpp", - "scene_name": "f8d5147d1d", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the power strip in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The power strip is to the side of and below the hole puncher, about 20 cm away. It is also to the side of and below the monitor, about 25 cm away. It is further to the side of and below the stapler, about 34 cm away." - }, - { - "id": "f8d5147d1d_3", - "dataset": "scannetpp", - "scene_name": "f8d5147d1d", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the power strip in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"power strip\", \"support_surface\": \"office table\", \"references\": [{\"object\": \"hole puncher\", \"relation\": \"side_below\", \"distance_cm\": 20.0}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 24.5}]}" - }, - { - "id": "f8d5147d1d_4", - "dataset": "scannetpp", - "scene_name": "f8d5147d1d", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the earphones appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "f8d5147d1d_5", - "dataset": "scannetpp", - "scene_name": "f8d5147d1d", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the stapler in this scene?\nA. monitor\nB. mouse\nC. office table\nD. cabinet", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. mouse" - }, - { - "id": "f8d5147d1d_6", - "dataset": "scannetpp", - "scene_name": "f8d5147d1d", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the stapler in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The stapler is next to the mouse, about 12 cm away. It is also to the side of and below the monitor, about 22 cm away. It is further to the side of and above the cabinet, about 27 cm away." - }, - { - "id": "f8d5147d1d_7", - "dataset": "scannetpp", - "scene_name": "f8d5147d1d", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the stapler in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"stapler\", \"support_surface\": \"office table\", \"references\": [{\"object\": \"mouse\", \"relation\": \"next_to\", \"distance_cm\": 12.1}, {\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 22.1}]}" - }, - { - "id": "f8f12e4e6b_0", - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the tv controller appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "f8f12e4e6b_1", - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. office chair\nB. box\nC. split\nD. shirt", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. box" - }, - { - "id": "f8f12e4e6b_2", - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is on the table. It is also next to the box, about 3 cm away. It is further next to the split, about 14 cm away." - }, - { - "id": "f8f12e4e6b_3", - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"table\", \"references\": [{\"object\": \"box\", \"relation\": \"next_to\", \"distance_cm\": 2.7}, {\"object\": \"split\", \"relation\": \"next_to\", \"distance_cm\": 13.8}]}" - }, - { - "id": "f8f12e4e6b_4", - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the tissue appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "f8f12e4e6b_5", - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tissue in this scene?\nA. chair\nB. shirt\nC. keyboard\nD. box", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. chair" - }, - { - "id": "f8f12e4e6b_6", - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tissue in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tissue is on the table. It is also to the side of and above the chair, about 16 cm away. It is further near the shirt, about 19 cm away." - }, - { - "id": "f8f12e4e6b_7", - "dataset": "scannetpp", - "scene_name": "f8f12e4e6b", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tissue in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tissue\", \"support_surface\": \"table\", \"references\": [{\"object\": \"chair\", \"relation\": \"side_above\", \"distance_cm\": 15.9}, {\"object\": \"shirt\", \"relation\": \"near\", \"distance_cm\": 18.6}]}" - }, - { - "id": "fb05e13ad1_0", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the tote bag appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "fb05e13ad1_1", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the backpack in this scene?\nA. book\nB. chair\nC. cabinet\nD. cushion", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "D. cushion" - }, - { - "id": "fb05e13ad1_2", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the backpack in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The backpack is to the side of and below the cushion, about 32 cm away." - }, - { - "id": "fb05e13ad1_3", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the backpack in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"backpack\", \"support_surface\": \"sofa chair\", \"references\": [{\"object\": \"cushion\", \"relation\": \"side_below\", \"distance_cm\": 31.9}]}" - }, - { - "id": "fb05e13ad1_4", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the cup appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "fb05e13ad1_5", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the cup in this scene?\nA. paper\nB. laptop\nC. book\nD. chair", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. book" - }, - { - "id": "fb05e13ad1_6", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the cup in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The cup is on the table. It is also to the side of and above the book, about 11 cm away. It is further to the side of and above the laptop, about 14 cm away." - }, - { - "id": "fb05e13ad1_7", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the cup in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"cup\", \"support_surface\": \"table\", \"references\": [{\"object\": \"book\", \"relation\": \"side_above\", \"distance_cm\": 10.8}, {\"object\": \"laptop\", \"relation\": \"side_above\", \"distance_cm\": 14.2}]}" - }, - { - "id": "fb05e13ad1_8", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "8", - "question_type": "target presence verification", - "question": "In the entire scene, did the mouse pad appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "fb05e13ad1_9", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "9", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the keyboard in this scene?\nA. monitor\nB. power socket\nC. book\nD. telephone", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "A. monitor" - }, - { - "id": "fb05e13ad1_10", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "10", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the keyboard in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The keyboard is to the side of and below the monitor, about 6 cm away. It is also next to the book, about 9 cm away. It is further to the side of and below the telephone, about 9 cm away." - }, - { - "id": "fb05e13ad1_11", - "dataset": "scannetpp", - "scene_name": "fb05e13ad1", - "sample_id": "11", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the keyboard in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"keyboard\", \"support_surface\": \"book\", \"references\": [{\"object\": \"monitor\", \"relation\": \"side_below\", \"distance_cm\": 5.7}, {\"object\": \"book\", \"relation\": \"next_to\", \"distance_cm\": 9.1}]}" - }, - { - "id": "fe1733741f_0", - "dataset": "scannetpp", - "scene_name": "fe1733741f", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the phone appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "fe1733741f_1", - "dataset": "scannetpp", - "scene_name": "fe1733741f", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone in this scene?\nA. plant\nB. books\nC. keyboard\nD. plant pot", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. books" - }, - { - "id": "fe1733741f_2", - "dataset": "scannetpp", - "scene_name": "fe1733741f", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone is to the side of and below the window, about 6 cm away. It is also next to the books, about 7 cm away. It is further near the keyboard, about 27 cm away." - }, - { - "id": "fe1733741f_3", - "dataset": "scannetpp", - "scene_name": "fe1733741f", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone\", \"support_surface\": \"window\", \"references\": [{\"object\": \"window\", \"relation\": \"side_below\", \"distance_cm\": 5.6}, {\"object\": \"books\", \"relation\": \"next_to\", \"distance_cm\": 7.0}]}" - }, - { - "id": "fe1733741f_4", - "dataset": "scannetpp", - "scene_name": "fe1733741f", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the toothbrush appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "fe1733741f_5", - "dataset": "scannetpp", - "scene_name": "fe1733741f", - "sample_id": "5", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the tote bag in this scene?\nA. mug\nB. cabinet\nC. box\nD. books", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. cabinet" - }, - { - "id": "fe1733741f_6", - "dataset": "scannetpp", - "scene_name": "fe1733741f", - "sample_id": "6", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the tote bag in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The tote bag is to the side of and below the cabinet, about 5 cm away. It is also to the side of and below the box, about 42 cm away. It is further to the side of and below the mug, about 71 cm away." - }, - { - "id": "fe1733741f_7", - "dataset": "scannetpp", - "scene_name": "fe1733741f", - "sample_id": "7", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the tote bag in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"tote bag\", \"support_surface\": \"cabinet\", \"references\": [{\"object\": \"cabinet\", \"relation\": \"side_below\", \"distance_cm\": 5.3}, {\"object\": \"box\", \"relation\": \"side_below\", \"distance_cm\": 42.0}]}" - }, - { - "id": "fe3fc057a1_0", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "0", - "question_type": "target presence verification", - "question": "In the entire scene, did the pencil case appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "fe3fc057a1_1", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "1", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the mouse in this scene?\nA. documents\nB. mug\nC. windowsill\nD. heater", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "C. windowsill" - }, - { - "id": "fe3fc057a1_2", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "2", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mouse in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mouse is on the table. It is also to the side of and below the windowsill, about 6 cm away. It is further to the side of and above the heater, about 17 cm away." - }, - { - "id": "fe3fc057a1_3", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "3", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mouse in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mouse\", \"support_surface\": \"table\", \"references\": [{\"object\": \"windowsill\", \"relation\": \"side_below\", \"distance_cm\": 5.6}, {\"object\": \"heater\", \"relation\": \"side_above\", \"distance_cm\": 17.2}]}" - }, - { - "id": "fe3fc057a1_4", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "4", - "question_type": "target presence verification", - "question": "In the entire scene, did the mug appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "Yes" - }, - { - "id": "fe3fc057a1_5", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "5", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the mug in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The mug is on the table. It is also to the side of and above the heater, about 6 cm away. It is further to the side of and below the windowsill, about 6 cm away." - }, - { - "id": "fe3fc057a1_6", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "6", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the mug in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"mug\", \"support_surface\": \"table\", \"references\": [{\"object\": \"heater\", \"relation\": \"side_above\", \"distance_cm\": 5.8}, {\"object\": \"windowsill\", \"relation\": \"side_below\", \"distance_cm\": 6.1}]}" - }, - { - "id": "fe3fc057a1_7", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "7", - "question_type": "target presence verification", - "question": "In the entire scene, did the lanyard appear?", - "instruction": "Answer only with exactly one word: Yes or No. Do not add any explanation.", - "ground_truth": "No" - }, - { - "id": "fe3fc057a1_8", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "8", - "question_type": "nearest reference identification", - "question": "Excluding the surface it is resting on, which of the following objects is nearest to the final location of the phone charger in this scene?\nA. kitchen cabinet\nB. stovetop\nC. frying pan\nD. kitchen shelf", - "instruction": "Choose exactly one option from A B C D.\nOutput ONLY in the format: X. object_name\nDo not output any other text.", - "ground_truth": "B. stovetop" - }, - { - "id": "fe3fc057a1_9", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "9", - "question_type": "fine-grained spatial description", - "question": "Please describe in detail the final location of the phone charger in the scene, including its distance and relation to nearby objects.", - "instruction": "Describe the location naturally in 1-2 sentences. Mention the support surface and 1-2 closest objects with exact distances in centimeters (e.g., 23 cm, 41 cm). Do not use meters or vague approximations. Use everyday English and varied sentence structures.", - "ground_truth": "The phone charger is to the side of and below the stovetop, about 26 cm away. It is also to the side of and below the kitchen shelf, about 39 cm away. It is further to the side of and below the frying pan, about 39 cm away." - }, - { - "id": "fe3fc057a1_10", - "dataset": "scannetpp", - "scene_name": "fe3fc057a1", - "sample_id": "10", - "question_type": "structured spatial prediction", - "question": "Please output the final location of the phone charger in structured JSON format.", - "instruction": "Output ONLY a valid JSON object. No extra text, no explanation, no markdown, no ```json.\n\nUse exactly these fields:\n- target: string (main object name)\n- support_surface: string (main surface or \"none\")\n- references: array of 0-2 objects, each with:\n - object: string\n - relation: one of on, under, above, below, side_above, side_below, next_to, near, attached_to\n - distance_cm: float (e.g. 8.0 or 20.4)\n\nExample:\n{\"target\": \"backpack\", \"support_surface\": \"none\", \"references\": [\n {\"object\": \"chair\", \"relation\": \"side_below\", \"distance_cm\": 17.0},\n {\"object\": \"nightstand\", \"relation\": \"side_below\", \"distance_cm\": 19.0}\n]}\n\nRules:\n1. Max 2 references, sorted by distance\n2. distance_cm must be float (add .0 if integer)\n3. JSON must be directly parsable with json.loads()", - "ground_truth": "{\"target\": \"phone charger\", \"support_surface\": \"kitchen shelf\", \"references\": [{\"object\": \"stovetop\", \"relation\": \"side_below\", \"distance_cm\": 25.7}, {\"object\": \"kitchen shelf\", \"relation\": \"side_below\", \"distance_cm\": 39.0}]}" - } - ] -} \ No newline at end of file