Datasets:
image image | question string | choices list | answer string | answer_type string | auxiliary_images images list | capability_name string | question_type_name string | difficulty string | id string | sample_id string | split string | capability_id string | question_type string | template_id string | template_variant_id string | option_labels list | option_texts list | options_json string | answer_format string | metric_name string | metric_family string | scorer string | scoring_rule string | vlmeval_compatible bool | difficulty_reason string | gt_fields list | metadata_json string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Please determine the approximate range of the number of visible category types in this image. | [
"A. 13-20 classes",
"B. 8-12 categories",
"C. 4-7 classes",
"D. 0-3 classes"
] | A | interval_choice | semantic category visibility | simple | TG_000041_Supermarket_P0000_000520_TG_C01_interval_choice_seg_count_interval_08 | Supermarket_P0000_000520 | test | TG_C01 | interval_choice | TG_C01_interval_choice_seg_count_interval_08 | TG_C01_interval_choice_seg_count_interval_08_v2 | [
"A",
"B",
"C",
"D"
] | [
"13-20 classes",
"8-12 categories",
"4-7 classes",
"0-3 classes"
] | {"A":"13-20 classes","B":"8-12 categories","C":"4-7 classes","D":"0-3 classes"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only count the number of visible categories in a single view and map it to a discrete interval, with evidence chain length of 1. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"categories":["beerbottle","building","cashregister","chairs","cucumbers","foilballoon","frozenfoodbagged","fryingoil","lightbulbstring","milkbottle","petfoodtub","sidewalk","walllight","woodenpallet"],"interval":"13-20 classes","visible_category_count":14},"answer_margin":14.0,"gt_lineage":{"answer... | |||
Please select multiple semantic region categories visible in the image. | [
"A. truck area",
"B. car area",
"C. road area",
"D. fence area",
"E. table area"
] | ["A","B","C"] | multi_choice | semantic category visibility | medium | TG_000006_ModernCityDowntown_P0000_000274_TG_C01_multi_choice_seg_category_multi_10 | ModernCityDowntown_P0000_000274 | test | TG_C01 | multi_choice | TG_C01_multi_choice_seg_category_multi_10 | TG_C01_multi_choice_seg_category_multi_10_v3 | [
"A",
"B",
"C",
"D",
"E"
] | [
"truck area",
"car area",
"road area",
"fence area",
"table area"
] | {"A":"truck area","B":"car area","C":"road area","D":"fence area","E":"table area"} | list | Set Exact Match + F1 | Exact Match / precision / recall / F1 | tartanground_gt_objective_metrics.set_scores | Compare predicted option-letter set with the GT option set; also report precision, recall and F1. | true | All visible categories must be selected from multiple candidates, with an evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"absent_distractors":["table","fence"],"visible_categories":["car","road","truck"],"visible_pixels":{"car":2248,"road":11822,"truck":8180}},"answer_margin":2248.0,"gt_lineage":{"answer_margin":2248.0,"difficulty":"medium","difficulty_reason":"All visible categories must be selected from multiple can... | |||
Which option corresponds to the semantic regions visible in the current image? | [
"A. planks area",
"B. crate area",
"C. metalchain area",
"D. bp area"
] | D | single_choice | semantic category visibility | simple | TG_000002_SeasideTown_P0000_000486_TG_C01_single_choice_seg_category_single_09 | SeasideTown_P0000_000486 | test | TG_C01 | single_choice | TG_C01_single_choice_seg_category_single_09 | TG_C01_single_choice_seg_category_single_09_v3 | [
"A",
"B",
"C",
"D"
] | [
"planks area",
"crate area",
"metalchain area",
"bp area"
] | {"A":"planks area","B":"crate area","C":"metalchain area","D":"bp area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only relying on the semantic segmentation category visibility in a single view, the evidence chain length is 1. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_category":"bp","distractors":["crate","metalchain","planks"],"visible_pixels":8655},"answer_margin":8655.0,"gt_lineage":{"answer_margin":8655.0,"difficulty":"simple","difficulty_reason":"Only relying on the semantic segmentation category visibility in a single view, the evidence chain length... | |||
Arrange the following categories in descending order of visible area size. | [
"A. streetlight area",
"B. building area",
"C. sidewalk area",
"D. road area"
] | ["D","B","C","A"] | ordered_list | area comparison of semantic regions | medium | TG_000005_ModernCityDowntown_P0000_000616_TG_C02_ordered_list_seg_area_order_04 | ModernCityDowntown_P0000_000616 | test | TG_C02 | ordered_list | TG_C02_ordered_list_seg_area_order_04 | TG_C02_ordered_list_seg_area_order_04_v1 | [
"A",
"B",
"C",
"D"
] | [
"streetlight area",
"building area",
"sidewalk area",
"road area"
] | {"A":"streetlight area","B":"building area","C":"sidewalk area","D":"road area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Sort the four semantic categories by visible pixel area, with evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_order_categories":["road","building","sidewalk","streetlight"],"area_by_category":{"building":129045,"road":168249,"sidewalk":22621,"streetlight":2303}},"answer_margin":20318.0,"gt_lineage":{"answer_margin":20318.0,"difficulty":"medium","difficulty_reason":"Sort the four semantic categories ... | |||
Which candidate area has the largest visible area? | [
"A. beam area",
"B. pillar area",
"C. rock area",
"D. supportbeam area"
] | D | single_choice | area comparison of semantic regions | medium | TG_000015_CoalMine_P0000_000512_TG_C02_single_choice_seg_area_largest_06 | CoalMine_P0000_000512 | test | TG_C02 | single_choice | TG_C02_single_choice_seg_area_largest_06 | TG_C02_single_choice_seg_area_largest_06_v4 | [
"A",
"B",
"C",
"D"
] | [
"beam area",
"pillar area",
"rock area",
"supportbeam area"
] | {"A":"beam area","B":"pillar area","C":"rock area","D":"supportbeam area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible pixel areas for multiple semantic categories, with evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_category":"supportbeam","area_by_category":{"beam":36811,"pillar":26715,"rock":2320,"supportbeam":134333}},"answer_margin":97522.0,"gt_lineage":{"answer_margin":97522.0,"difficulty":"medium","difficulty_reason":"Compare the visible pixel areas for multiple semantic categories, with evidence ... | |||
If only one is chosen from the A-I region, where is the groundembankment region most concentrated? | [
"A. Label region D",
"B. Annotation area F",
"C. E annotated area",
"D. Label area A"
] | B | grounding_choice | image region semantic localization | medium | TG_000004_ForestEnv_P0000_001183_TG_C03_grounding_choice_grid_category_location_01 | ForestEnv_P0000_001183 | test | TG_C03 | grounding_choice | TG_C03_grounding_choice_grid_category_location_01 | TG_C03_grounding_choice_grid_category_location_01_v1 | [
"A",
"B",
"C",
"D"
] | [
"Label region D",
"Annotation area F",
"E annotated area",
"Label area A"
] | {"A":"Label region D","B":"Annotation area F","C":"E annotated area","D":"Label area A"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Aggregate semantic category pixels into a 3x3 grid and select the region with the largest area; the evidence chain length is 2. | [
"image_<camera>/*.png",
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_region":"F","area_by_region":{"A":0,"D":1653,"E":100,"F":6540},"target_category":"groundembankment"},"answer_margin":4887.0,"gt_lineage":{"answer_margin":4887.0,"difficulty":"medium","difficulty_reason":"Aggregate semantic category pixels into a 3x3 grid and select the region with the larges... | |||
Please provide the corresponding category option letters for the specified regions. The regions to match are: E, B, D. | [
"A. pipeline area",
"B. airconditioner area",
"C. barrel area"
] | {"B":"A","D":"A","E":"A"} | matching | image region semantic localization | hard | TG_000008_CoalMine_P0000_000000_TG_C03_matching_grid_region_category_matching_08 | CoalMine_P0000_000000 | test | TG_C03 | matching | TG_C03_matching_grid_region_category_matching_08 | TG_C03_matching_grid_region_category_matching_08_v1 | [
"A",
"B",
"C"
] | [
"pipeline area",
"airconditioner area",
"barrel area"
] | {"A":"pipeline area","B":"airconditioner area","C":"barrel area"} | mapping | DictExactStrMatchAggRecall | structured exact matching | tartanground_gt_objective_metrics.matching_scores | Parse a mapping and compare every key-value pair exactly. | true | Simultaneously determine the dominant category for three annotated regions and output the mapping, with evidence chain length of 3. | [
"image_<camera>/*.png",
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"category_options":{"A":"pipeline area","B":"airconditioner area","C":"barrel area"},"region_to_category":{"B":"pipe","D":"pipe","E":"pipe"}},"answer_margin":25266.0,"gt_lineage":{"answer_margin":25266.0,"difficulty":"hard","difficulty_reason":"Simultaneously determine the dominant category for thre... | |||
Which option corresponds to the main content of area D? | [
"A. tree area",
"B. ground area",
"C. truckbed area",
"D. utilitypole area"
] | A | single_choice | image region semantic localization | simple | TG_006622_GreatMarsh_P0000_000710_TG_C03_single_choice_grid_region_main_class_09 | GreatMarsh_P0000_000710 | test | TG_C03 | single_choice | TG_C03_single_choice_grid_region_main_class_09 | TG_C03_single_choice_grid_region_main_class_09_v1 | [
"A",
"B",
"C",
"D"
] | [
"tree area",
"ground area",
"truckbed area",
"utilitypole area"
] | {"A":"tree area","B":"ground area","C":"truckbed area","D":"utilitypole area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Only judge the dominant semantic category in one annotated region, with evidence chain length of 1. | [
"image_<camera>/*.png",
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_category":"tree","category_area_by_option":{"ground":13749,"tree":16729,"truckbed":5114,"utilitypole":508},"region":"D"},"answer_margin":2980.0,"gt_lineage":{"answer_margin":2980.0,"difficulty":"simple","difficulty_reason":"Only judge the dominant semantic category in one annotated region, w... | |||
Please determine which labeled area is closer in position. | [
"A. Label region D",
"B. Annotation area H",
"C. G annotated area",
"D. I labeling area"
] | D | grounding_choice | distance and depth ordering | medium | TG_000034_Office_P0000_000606_TG_C04_grounding_choice_depth_region_nearest_02 | Office_P0000_000606 | test | TG_C04 | grounding_choice | TG_C04_grounding_choice_depth_region_nearest_02 | TG_C04_grounding_choice_depth_region_nearest_02_v4 | [
"A",
"B",
"C",
"D"
] | [
"Label region D",
"Annotation area H",
"G annotated area",
"I labeling area"
] | {"A":"Label region D","B":"Annotation area H","C":"G annotated area","D":"I labeling area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the effective median distances of multiple annotated areas, with evidence chain length of 2. | [
"image_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"answer_region":"I","median_depth_by_region":{"D":2.01175,"G":2.00195,"H":2.00195,"I":0.50245}},"answer_margin":1.4995012879371643,"gt_lineage":{"answer_margin":1.4995012879371643,"difficulty":"medium","difficulty_reason":"Compare the effective median distances of multiple annotated areas, with evid... | |||
Only considering area B, which range is its overall proximity-farther more similar to? | [
"A. 20 meters and above",
"B. 1-3 meters",
"C. 8-20 meters",
"D. less than 1 meter",
"E. 3-8 meters"
] | D | interval_choice | distance and depth ordering | simple | TG_000001_CoalMine_P0000_000695_TG_C04_interval_choice_depth_region_interval_08 | CoalMine_P0000_000695 | test | TG_C04 | interval_choice | TG_C04_interval_choice_depth_region_interval_08 | TG_C04_interval_choice_depth_region_interval_08_v1 | [
"A",
"B",
"C",
"D",
"E"
] | [
"20 meters and above",
"1-3 meters",
"8-20 meters",
"less than 1 meter",
"3-8 meters"
] | {"A":"20 meters and above","B":"1-3 meters","C":"8-20 meters","D":"less than 1 meter","E":"3-8 meters"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single annotated area's valid median proximity mapping to discrete interval, evidence chain length of 1. | [
"image_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"interval":"less than 1 meter","median_depth":0.50196,"region":"B","valid_fraction":1.0},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"simple","difficulty_reason":"Single annotated area's valid median proximity mapping to discrete interval, evidence chain length of 1.","distrac... | |||
Which range are the overall front-to-back distances of the foliage area in the image closer to? | [
"A. 8-20 meters",
"B. 1-3 meters",
"C. less than 1 meter",
"D. 3-8 meters",
"E. 20 meters and above"
] | A | interval_choice | semantic category and combined judgment of proximity and distance | medium | TG_000016_ForestEnv_P0000_000631_TG_C05_interval_choice_semantic_depth_interval_02 | ForestEnv_P0000_000631 | test | TG_C05 | interval_choice | TG_C05_interval_choice_semantic_depth_interval_02 | TG_C05_interval_choice_semantic_depth_interval_02_v1 | [
"A",
"B",
"C",
"D",
"E"
] | [
"8-20 meters",
"1-3 meters",
"less than 1 meter",
"3-8 meters",
"20 meters and above"
] | {"A":"8-20 meters","B":"1-3 meters","C":"less than 1 meter","D":"3-8 meters","E":"20 meters and above"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | After applying a semantic mask to constrain the data, the near-far interval was counted, resulting in an evidence chain length of 2. | [
"seg_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"category":"foliage","interval":"8-20 meters","median_depth":8.02361,"valid_fraction":1.0},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"medium","difficulty_reason":"After applying a semantic mask to constrain the data, the near-far interval was counted, resulting in an evidenc... | |||
Compare the overall distances of the candidate categories and rank them from nearest to farthest. | [
"A. walkway area",
"B. stairs area",
"C. water area",
"D. pipeline area"
] | ["A","D","B","C"] | ordered_list | semantic category and combined judgment of proximity and distance | hard | TG_000038_SeasideTown_P0000_000030_TG_C05_ordered_list_semantic_depth_order_01 | SeasideTown_P0000_000030 | test | TG_C05 | ordered_list | TG_C05_ordered_list_semantic_depth_order_01 | TG_C05_ordered_list_semantic_depth_order_01_v4 | [
"A",
"B",
"C",
"D"
] | [
"walkway area",
"stairs area",
"water area",
"pipeline area"
] | {"A":"walkway area","B":"stairs area","C":"water area","D":"pipeline area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Sort the effective median distances for multiple semantic masks, with evidence chain length of 3. | [
"seg_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"answer_order_categories":["walkway","pipe","stairs","water"],"median_depth_by_category":{"pipe":2.00787,"stairs":8.03131,"walkway":0.50196,"water":32.15657}},"answer_margin":1.5059058666229248,"gt_lineage":{"answer_margin":1.5059058666229248,"difficulty":"hard","difficulty_reason":"Sort the effecti... | |||
Which candidate category is closer in semantic space? | [
"A. wall area",
"B. building area",
"C. atm area",
"D. sidewalk area"
] | D | single_choice | semantic category and combined judgment of proximity and distance | hard | TG_000010_ModernCityDowntown_P0000_000479_TG_C05_single_choice_semantic_depth_nearest_05 | ModernCityDowntown_P0000_000479 | test | TG_C05 | single_choice | TG_C05_single_choice_semantic_depth_nearest_05 | TG_C05_single_choice_semantic_depth_nearest_05_v5 | [
"A",
"B",
"C",
"D"
] | [
"wall area",
"building area",
"atm area",
"sidewalk area"
] | {"A":"wall area","B":"building area","C":"atm area","D":"sidewalk area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Simultaneously relying on semantic mask and distance statistics for category comparison, with an evidence chain length of 3. | [
"seg_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"answer_category":"sidewalk","median_depth_by_category":{"atm":8.02363,"building":8.03148,"sidewalk":0.50294,"wall":32.09428}},"answer_margin":7.520692765712738,"gt_lineage":{"answer_margin":7.520692765712738,"difficulty":"hard","difficulty_reason":"Simultaneously relying on semantic mask and distan... | |||
Observe V/W/X/Y, and match them to their main categories. Views to be matched: V, W, X, Y. | [
"A. walls area",
"B. airconditioner area",
"C. babybed area"
] | {"V":"A","W":"A","X":"A","Y":"A"} | matching | multi-view semantic consistency | hard | TG_000017_House_P0000_000168_TG_C06_matching_multiview_dominant_matching_02 | House_P0000_000168 | test | TG_C06 | matching | TG_C06_matching_multiview_dominant_matching_02 | TG_C06_matching_multiview_dominant_matching_02_v1 | [
"A",
"B",
"C"
] | [
"walls area",
"airconditioner area",
"babybed area"
] | {"A":"walls area","B":"airconditioner area","C":"babybed area"} | mapping | DictExactStrMatchAggRecall | structured exact matching | tartanground_gt_objective_metrics.matching_scores | Parse a mapping and compare every key-value pair exactly. | true | Simultaneously matching multiple synchronized views with the main semantic category, evidence chain length of 3. | [
"multi-camera image_<camera>",
"multi-camera seg_<camera>",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"category_options":{"A":"walls area","B":"airconditioner area","C":"babybed area"},"view_to_category":{"V":"walls","W":"walls","X":"walls","Y":"walls"}},"answer_margin":49321.0,"gt_lineage":{"answer_margin":49321.0,"difficulty":"hard","difficulty_reason":"Simultaneously matching multiple synchronize... | |||
Select the option with the highest proportion of the sign region from the marked view. | [
"A. Labeled View V",
"B. Labeled View W",
"C. Label view X",
"D. annotate view Y"
] | C | multi_view_choice | multi-view semantic consistency | hard | TG_000007_JapaneseAlley_P0000_000792_TG_C06_multi_view_choice_multiview_category_area_08 | JapaneseAlley_P0000_000792 | test | TG_C06 | multi_view_choice | TG_C06_multi_view_choice_multiview_category_area_08 | TG_C06_multi_view_choice_multiview_category_area_08_v1 | [
"A",
"B",
"C",
"D"
] | [
"Labeled View V",
"Labeled View W",
"Label view X",
"annotate view Y"
] | {"A":"Labeled View V","B":"Labeled View W","C":"Label view X","D":"annotate view Y"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible area of the same semantic category across four synchronized viewpoints, with evidence chain length of 3. | [
"multi-camera image_<camera>",
"multi-camera seg_<camera>",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_view":"X","area_by_view":{"V":0,"W":0,"X":9846,"Y":4069},"target_category":"sign"},"answer_margin":5777.0,"gt_lineage":{"answer_margin":5777.0,"difficulty":"hard","difficulty_reason":"Compare the visible area of the same semantic category across four synchronized viewpoints, with evidence ch... | |||
What is the order in which these annotated points appear as the path is traversed? | [
"A. Annotation point A",
"B. label B",
"C. label C",
"D. Label D"
] | ["C","A","D","B"] | ordered_list | trajectory order and path structure | hard | TG_000009_Hospital_P0000_001183_TG_C07_ordered_list_trajectory_point_order_06 | Hospital_P0000_001183 | test | TG_C07 | ordered_list | TG_C07_ordered_list_trajectory_point_order_06 | TG_C07_ordered_list_trajectory_point_order_06_v4 | [
"A",
"B",
"C",
"D"
] | [
"Annotation point A",
"label B",
"label C",
"Label D"
] | {"A":"Annotation point A","B":"label B","C":"label C","D":"Label D"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Source text needs to be sorted in the direction of the GT path, with multiple scrambled trajectory points, and the evidence chain length is 3. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"frame_ids_in_order":[394,473,552,631],"ordered_point_labels":["C","A","D","B"]},"answer_margin":0.4,"gt_lineage":{"answer_margin":0.4,"difficulty":"hard","difficulty_reason":"Source text needs to be sorted in the direction of the GT path, with multiple scrambled trajectory points, and the evidence ... | |||
Observe the S-G trajectory, which phase interval does A belong to? | [
"A. middle segment",
"B. previous segment",
"C. back segment"
] | B | temporal_choice | trajectory order and path structure | simple | TG_000012_Downtown_P0000_000285_TG_C07_temporal_choice_trajectory_stage_interval_05 | Downtown_P0000_000285 | test | TG_C07 | temporal_choice | TG_C07_temporal_choice_trajectory_stage_interval_05 | TG_C07_temporal_choice_trajectory_stage_interval_05_v1 | [
"A",
"B",
"C"
] | [
"middle segment",
"previous segment",
"back segment"
] | {"A":"middle segment","B":"previous segment","C":"back segment"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Sequence position of individual path points mapped to front/middle/back stages, with evidence chain length of 1. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"frame_id":71,"normalized_rank":0.05263,"point_label":"A","stage":"previous segment"},"answer_margin":0.4473684210526316,"gt_lineage":{"answer_margin":0.4473684210526316,"difficulty":"simple","difficulty_reason":"Sequence position of individual path points mapped to front/middle/back stages, with ev... | |||
How should the displacement scales of these segments be ranked? | [
"A. Labelled Road A",
"B. Annotated segment B",
"C. labeled road segment C",
"D. Label: D"
] | ["A","C","D","B"] | ordered_list | motion trend and orientation change | hard | TG_000019_OldIndustrialCity_P0000_000000_TG_C08_ordered_list_motion_segment_speed_order_06 | OldIndustrialCity_P0000_000000 | test | TG_C08 | ordered_list | TG_C08_ordered_list_motion_segment_speed_order_06 | TG_C08_ordered_list_motion_segment_speed_order_06_v2 | [
"A",
"B",
"C",
"D"
] | [
"Labelled Road A",
"Annotated segment B",
"labeled road segment C",
"Label: D"
] | {"A":"Labelled Road A","B":"Annotated segment B","C":"labeled road segment C","D":"Label: D"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Use segment displacement length as a proxy for speed and sort at equal time intervals, with evidence chain length of 3. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"answer_order_segments":["A","C","D","B"],"segment_length_by_label":{"A":23.00581,"B":1.73988,"C":13.28436,"D":9.915}},"answer_margin":3.3693580995002463,"gt_lineage":{"answer_margin":3.3693580995002463,"difficulty":"hard","difficulty_reason":"Use segment displacement length as a proxy for speed and... | |||
Observe the direction of the trajectory changes. Which side is this path mainly turning towards? | [
"A. Basic Straight Run",
"B. mainly turn left",
"C. mainly turn right"
] | B | single_choice | motion trend and orientation change | medium | TG_000025_JapaneseAlley_P0000_000288_TG_C08_single_choice_motion_turn_choice_07 | JapaneseAlley_P0000_000288 | test | TG_C08 | single_choice | TG_C08_single_choice_motion_turn_choice_07 | TG_C08_single_choice_motion_turn_choice_07_v3 | [
"A",
"B",
"C"
] | [
"Basic Straight Run",
"mainly turn left",
"mainly turn right"
] | {"A":"Basic Straight Run","B":"mainly turn left","C":"mainly turn right"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Determine the turning trend based on the change in start-end orientation angle, with evidence chain length of 2. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"turn_class":"mainly turn left","yaw_delta_deg":29.21945},"answer_margin":29.219452922077874,"gt_lineage":{"answer_margin":29.219452922077874,"difficulty":"medium","difficulty_reason":"Determine the turning trend based on the change in start-end orientation angle, with evidence chain length of 2.","... | |||
Observe the trajectory graph; which item does this path's overall movement direction align more closely with? | [
"A. southwest",
"B. west",
"C. towards northwest",
"D. North",
"E. East",
"F. Southeast",
"G. South",
"H. northeast"
] | A | temporal_choice | motion trend and orientation change | medium | TG_000003_ForestEnv_P0000_000394_TG_C08_temporal_choice_motion_direction_choice_07 | ForestEnv_P0000_000394 | test | TG_C08 | temporal_choice | TG_C08_temporal_choice_motion_direction_choice_07 | TG_C08_temporal_choice_motion_direction_choice_07_v3 | [
"A",
"B",
"C",
"D",
"E",
"F",
"G",
"H"
] | [
"southwest",
"west",
"towards northwest",
"North",
"East",
"Southeast",
"South",
"northeast"
] | {"A":"southwest","B":"west","C":"towards northwest","D":"North","E":"East","F":"Southeast","G":"South","H":"northeast"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Based on the GT displacement between the start and end points, the overall direction is determined, with an evidence chain length of 2. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"direction":"southwest","net_displacement_xy":[-31.85628,-14.63146]},"answer_margin":35.05570071892439,"gt_lineage":{"answer_margin":35.05570071892439,"difficulty":"medium","difficulty_reason":"Based on the GT displacement between the start and end points, the overall direction is determined, with a... | |||
Please select all the actual area categories that appear in the candidates. | [
"A. mobilegenerator area",
"B. wheelbarrow area",
"C. floodlightpole area",
"D. wall area",
"E. rock area"
] | ["A","C","D"] | multi_choice | semantic category visibility | medium | TG_000013_ConstructionSite_P0000_000555_TG_C01_multi_choice_seg_category_multi_01 | ConstructionSite_P0000_000555 | test | TG_C01 | multi_choice | TG_C01_multi_choice_seg_category_multi_01 | TG_C01_multi_choice_seg_category_multi_01_v4 | [
"A",
"B",
"C",
"D",
"E"
] | [
"mobilegenerator area",
"wheelbarrow area",
"floodlightpole area",
"wall area",
"rock area"
] | {"A":"mobilegenerator area","B":"wheelbarrow area","C":"floodlightpole area","D":"wall area","E":"rock area"} | list | Set Exact Match + F1 | Exact Match / precision / recall / F1 | tartanground_gt_objective_metrics.set_scores | Compare predicted option-letter set with the GT option set; also report precision, recall and F1. | true | All visible categories must be selected from multiple candidates, with an evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"absent_distractors":["wheelbarrow","rock"],"visible_categories":["floodlightpole","mobilegenerator","wall"],"visible_pixels":{"floodlightpole":1118,"mobilegenerator":1275,"wall":1198}},"answer_margin":1118.0,"gt_lineage":{"answer_margin":1118.0,"difficulty":"medium","difficulty_reason":"All visible... | |||
Arrange the following categories in descending order of visible area size. | [
"A. wall area",
"B. water area",
"C. building area",
"D. bridge area"
] | ["D","C","A","B"] | ordered_list | area comparison of semantic regions | medium | TG_000026_NordicHarbor_P0000_001026_TG_C02_ordered_list_seg_area_order_08 | NordicHarbor_P0000_001026 | test | TG_C02 | ordered_list | TG_C02_ordered_list_seg_area_order_08 | TG_C02_ordered_list_seg_area_order_08_v4 | [
"A",
"B",
"C",
"D"
] | [
"wall area",
"water area",
"building area",
"bridge area"
] | {"A":"wall area","B":"water area","C":"building area","D":"bridge area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Sort the four semantic categories by visible pixel area, with evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_order_categories":["bridge","building","wall","water"],"area_by_category":{"bridge":159660,"building":86519,"wall":13339,"water":8704}},"answer_margin":4635.0,"gt_lineage":{"answer_margin":4635.0,"difficulty":"medium","difficulty_reason":"Sort the four semantic categories by visible pixel ar... | |||
Select the category with the largest visible field of view in the current image. | [
"A. stonedebris area",
"B. chimney area",
"C. shipconstruction area",
"D. building area"
] | D | single_choice | area comparison of semantic regions | medium | TG_000036_OldIndustrialCity_P0000_000756_TG_C02_single_choice_seg_area_largest_02 | OldIndustrialCity_P0000_000756 | test | TG_C02 | single_choice | TG_C02_single_choice_seg_area_largest_02 | TG_C02_single_choice_seg_area_largest_02_v5 | [
"A",
"B",
"C",
"D"
] | [
"stonedebris area",
"chimney area",
"shipconstruction area",
"building area"
] | {"A":"stonedebris area","B":"chimney area","C":"shipconstruction area","D":"building area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible pixel areas for multiple semantic categories, with evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_category":"building","area_by_category":{"building":109713,"chimney":3680,"shipconstruction":1233,"stonedebris":66656}},"answer_margin":43057.0,"gt_lineage":{"answer_margin":43057.0,"difficulty":"medium","difficulty_reason":"Compare the visible pixel areas for multiple semantic categories, w... | |||
If only one is chosen in the A-I region, where is the ceiling area most concentrated? | [
"A. Label region D",
"B. Label area A",
"C. B annotated area",
"D. C labeled area"
] | B | grounding_choice | image region semantic localization | medium | TG_000014_Restaurant_P0000_000524_TG_C03_grounding_choice_grid_category_location_03 | Restaurant_P0000_000524 | test | TG_C03 | grounding_choice | TG_C03_grounding_choice_grid_category_location_03 | TG_C03_grounding_choice_grid_category_location_03_v1 | [
"A",
"B",
"C",
"D"
] | [
"Label region D",
"Label area A",
"B annotated area",
"C labeled area"
] | {"A":"Label region D","B":"Label area A","C":"B annotated area","D":"C labeled area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Aggregate semantic category pixels into a 3x3 grid and select the region with the largest area; the evidence chain length is 2. | [
"image_<camera>/*.png",
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_region":"A","area_by_region":{"A":21652,"B":12667,"C":16542,"D":0},"target_category":"ceiling"},"answer_margin":5110.0,"gt_lineage":{"answer_margin":5110.0,"difficulty":"medium","difficulty_reason":"Aggregate semantic category pixels into a 3x3 grid and select the region with the largest are... | |||
Please match the labeled areas with the candidate category options. Areas to match: A, F, E. | [
"A. building area",
"B. door area",
"C. column Area"
] | {"A":"A","E":"C","F":"B"} | matching | image region semantic localization | hard | TG_000029_OldIndustrialCity_P0000_000403_TG_C03_matching_grid_region_category_matching_07 | OldIndustrialCity_P0000_000403 | test | TG_C03 | matching | TG_C03_matching_grid_region_category_matching_07 | TG_C03_matching_grid_region_category_matching_07_v1 | [
"A",
"B",
"C"
] | [
"building area",
"door area",
"column Area"
] | {"A":"building area","B":"door area","C":"column Area"} | mapping | DictExactStrMatchAggRecall | structured exact matching | tartanground_gt_objective_metrics.matching_scores | Parse a mapping and compare every key-value pair exactly. | true | Simultaneously determine the dominant category for three annotated regions and output the mapping, with evidence chain length of 3. | [
"image_<camera>/*.png",
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"category_options":{"A":"building area","B":"door area","C":"column Area"},"region_to_category":{"A":"building","E":"column","F":"door"}},"answer_margin":6575.0,"gt_lineage":{"answer_margin":6575.0,"difficulty":"hard","difficulty_reason":"Simultaneously determine the dominant category for three anno... | |||
Observe the annotated image and determine which annotated area is overall closer to the observer. | [
"A. Annotation area F",
"B. C labeled area",
"C. G annotated area",
"D. Label area A"
] | C | grounding_choice | distance and depth ordering | medium | TG_000043_CoalMine_P0000_000622_TG_C04_grounding_choice_depth_region_nearest_05 | CoalMine_P0000_000622 | test | TG_C04 | grounding_choice | TG_C04_grounding_choice_depth_region_nearest_05 | TG_C04_grounding_choice_depth_region_nearest_05_v4 | [
"A",
"B",
"C",
"D"
] | [
"Annotation area F",
"C labeled area",
"G annotated area",
"Label area A"
] | {"A":"Annotation area F","B":"C labeled area","C":"G annotated area","D":"Label area A"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the effective median distances of multiple annotated areas, with evidence chain length of 2. | [
"image_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"answer_region":"G","median_depth_by_region":{"A":2.0098,"C":2.00982,"F":2.00981,"G":0.50148}},"answer_margin":1.508329153060913,"gt_lineage":{"answer_margin":1.508329153060913,"difficulty":"medium","difficulty_reason":"Compare the effective median distances of multiple annotated areas, with evidenc... | |||
which interval does the typical visible position in region G belong to | [
"A. 1-3 meters",
"B. 8-20 meters",
"C. 20 meters and above",
"D. less than 1 meter",
"E. 3-8 meters"
] | D | interval_choice | distance and depth ordering | simple | TG_000011_ModernCityDowntown_P0000_000547_TG_C04_interval_choice_depth_region_interval_04 | ModernCityDowntown_P0000_000547 | test | TG_C04 | interval_choice | TG_C04_interval_choice_depth_region_interval_04 | TG_C04_interval_choice_depth_region_interval_04_v1 | [
"A",
"B",
"C",
"D",
"E"
] | [
"1-3 meters",
"8-20 meters",
"20 meters and above",
"less than 1 meter",
"3-8 meters"
] | {"A":"1-3 meters","B":"8-20 meters","C":"20 meters and above","D":"less than 1 meter","E":"3-8 meters"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single annotated area's valid median proximity mapping to discrete interval, evidence chain length of 1. | [
"image_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"interval":"less than 1 meter","median_depth":0.50196,"region":"G","valid_fraction":1.0},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"simple","difficulty_reason":"Single annotated area's valid median proximity mapping to discrete interval, evidence chain length of 1.","distrac... | |||
Select the typical distance range corresponding to the woodpole area. | [
"A. 8-20 meters",
"B. less than 1 meter",
"C. 1-3 meters",
"D. 3-8 meters",
"E. 20 meters and above"
] | B | interval_choice | semantic category and combined judgment of proximity and distance | medium | TG_000046_JapaneseAlley_P0000_000000_TG_C05_interval_choice_semantic_depth_interval_08 | JapaneseAlley_P0000_000000 | test | TG_C05 | interval_choice | TG_C05_interval_choice_semantic_depth_interval_08 | TG_C05_interval_choice_semantic_depth_interval_08_v1 | [
"A",
"B",
"C",
"D",
"E"
] | [
"8-20 meters",
"less than 1 meter",
"1-3 meters",
"3-8 meters",
"20 meters and above"
] | {"A":"8-20 meters","B":"less than 1 meter","C":"1-3 meters","D":"3-8 meters","E":"20 meters and above"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | After applying a semantic mask to constrain the data, the near-far interval was counted, resulting in an evidence chain length of 2. | [
"seg_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"category":"woodpole","interval":"less than 1 meter","median_depth":0.50196,"valid_fraction":1.0},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"medium","difficulty_reason":"After applying a semantic mask to constrain the data, the near-far interval was counted, resulting in an ... | |||
From closer to the observer to farther away, what is the class order? | [
"A. saucebottle area",
"B. fryingoil area",
"C. cashregister area",
"D. petfoodtub area"
] | ["C","D","A","B"] | ordered_list | semantic category and combined judgment of proximity and distance | hard | TG_000197_Supermarket_P0000_000560_TG_C05_ordered_list_semantic_depth_order_09 | Supermarket_P0000_000560 | test | TG_C05 | ordered_list | TG_C05_ordered_list_semantic_depth_order_09 | TG_C05_ordered_list_semantic_depth_order_09_v3 | [
"A",
"B",
"C",
"D"
] | [
"saucebottle area",
"fryingoil area",
"cashregister area",
"petfoodtub area"
] | {"A":"saucebottle area","B":"fryingoil area","C":"cashregister area","D":"petfoodtub area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Sort the effective median distances for multiple semantic masks, with evidence chain length of 3. | [
"seg_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"answer_order_categories":["cashregister","petfoodtub","saucebottle","fryingoil"],"median_depth_by_category":{"cashregister":0.50196,"fryingoil":32.12523,"petfoodtub":2.01177,"saucebottle":8.0313}},"answer_margin":1.5098055601119995,"gt_lineage":{"answer_margin":1.5098055601119995,"difficulty":"hard... | |||
Which category is the closest in typical visible distance? | [
"A. door area",
"B. trim area",
"C. shutter area",
"D. wall area"
] | B | single_choice | semantic category and combined judgment of proximity and distance | hard | TG_000028_SeasideTown_P0000_000121_TG_C05_single_choice_semantic_depth_nearest_10 | SeasideTown_P0000_000121 | test | TG_C05 | single_choice | TG_C05_single_choice_semantic_depth_nearest_10 | TG_C05_single_choice_semantic_depth_nearest_10_v4 | [
"A",
"B",
"C",
"D"
] | [
"door area",
"trim area",
"shutter area",
"wall area"
] | {"A":"door area","B":"trim area","C":"shutter area","D":"wall area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Simultaneously relying on semantic mask and distance statistics for category comparison, with an evidence chain length of 3. | [
"seg_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"answer_category":"trim","median_depth_by_category":{"door":8.02356,"shutter":8.02352,"trim":2.00784,"wall":8.01573}},"answer_margin":6.0078864097595215,"gt_lineage":{"answer_margin":6.0078864097595215,"difficulty":"hard","difficulty_reason":"Simultaneously relying on semantic mask and distance stat... | |||
Which category does each annotated view mainly belong to? Views to match: V, W, X, Y. | [
"A. wall area",
"B. ground area"
] | {"V":"A","W":"B","X":"A","Y":"A"} | matching | multi-view semantic consistency | hard | TG_000027_AbandonedFactory_P0000_001181_TG_C06_matching_multiview_dominant_matching_03 | AbandonedFactory_P0000_001181 | test | TG_C06 | matching | TG_C06_matching_multiview_dominant_matching_03 | TG_C06_matching_multiview_dominant_matching_03_v1 | [
"A",
"B"
] | [
"wall area",
"ground area"
] | {"A":"wall area","B":"ground area"} | mapping | DictExactStrMatchAggRecall | structured exact matching | tartanground_gt_objective_metrics.matching_scores | Parse a mapping and compare every key-value pair exactly. | true | Simultaneously matching multiple synchronized views with the main semantic category, evidence chain length of 3. | [
"multi-camera image_<camera>",
"multi-camera seg_<camera>",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"category_options":{"A":"wall area","B":"ground area"},"view_to_category":{"V":"wall","W":"ground","X":"wall","Y":"wall"}},"answer_margin":28613.0,"gt_lineage":{"answer_margin":28613.0,"difficulty":"hard","difficulty_reason":"Simultaneously matching multiple synchronized views with the main semantic... | |||
Compare the views, which one shows the most metalshelves area? | [
"A. Labeled View V",
"B. Labeled View W",
"C. Label view X",
"D. annotate view Y"
] | B | multi_view_choice | multi-view semantic consistency | hard | TG_000018_JapaneseCity_P0000_000279_TG_C06_multi_view_choice_multiview_category_area_07 | JapaneseCity_P0000_000279 | test | TG_C06 | multi_view_choice | TG_C06_multi_view_choice_multiview_category_area_07 | TG_C06_multi_view_choice_multiview_category_area_07_v1 | [
"A",
"B",
"C",
"D"
] | [
"Labeled View V",
"Labeled View W",
"Label view X",
"annotate view Y"
] | {"A":"Labeled View V","B":"Labeled View W","C":"Label view X","D":"annotate view Y"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible area of the same semantic category across four synchronized viewpoints, with evidence chain length of 3. | [
"multi-camera image_<camera>",
"multi-camera seg_<camera>",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_view":"W","area_by_view":{"V":3225,"W":7383,"X":0,"Y":0},"target_category":"metalshelves"},"answer_margin":4158.0,"gt_lineage":{"answer_margin":4158.0,"difficulty":"hard","difficulty_reason":"Compare the visible area of the same semantic category across four synchronized viewpoints, with evi... | |||
Pushing along the trajectory, to which phase should C be assigned? | [
"A. previous segment",
"B. back segment",
"C. middle segment"
] | B | temporal_choice | trajectory order and path structure | simple | TG_000021_GreatMarsh_P0000_001420_TG_C07_temporal_choice_trajectory_stage_interval_08 | GreatMarsh_P0000_001420 | test | TG_C07 | temporal_choice | TG_C07_temporal_choice_trajectory_stage_interval_08 | TG_C07_temporal_choice_trajectory_stage_interval_08_v1 | [
"A",
"B",
"C"
] | [
"previous segment",
"back segment",
"middle segment"
] | {"A":"previous segment","B":"back segment","C":"middle segment"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Sequence position of individual path points mapped to front/middle/back stages, with evidence chain length of 1. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"frame_id":1341,"normalized_rank":0.89474,"point_label":"C","stage":"back segment"},"answer_margin":0.39473684210526316,"gt_lineage":{"answer_margin":0.39473684210526316,"difficulty":"simple","difficulty_reason":"Sequence position of individual path points mapped to front/middle/back stages, with ev... | |||
Please sort the annotated segments in descending order of travel distance. | [
"A. Labelled Road A",
"B. Annotated segment B",
"C. labeled road segment C",
"D. Label: D"
] | ["A","B","C","D"] | ordered_list | motion trend and orientation change | hard | TG_000047_ConstructionSite_P0000_000392_TG_C08_ordered_list_motion_segment_speed_order_07 | ConstructionSite_P0000_000392 | test | TG_C08 | ordered_list | TG_C08_ordered_list_motion_segment_speed_order_07 | TG_C08_ordered_list_motion_segment_speed_order_07_v2 | [
"A",
"B",
"C",
"D"
] | [
"Labelled Road A",
"Annotated segment B",
"labeled road segment C",
"Label: D"
] | {"A":"Labelled Road A","B":"Annotated segment B","C":"labeled road segment C","D":"Label: D"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Use segment displacement length as a proxy for speed and sort at equal time intervals, with evidence chain length of 3. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"answer_order_segments":["A","B","C","D"],"segment_length_by_label":{"A":8.80806,"B":8.50912,"C":8.10834,"D":7.31669}},"answer_margin":0.29894033921591046,"gt_lineage":{"answer_margin":0.29894033921591046,"difficulty":"hard","difficulty_reason":"Use segment displacement length as a proxy for speed a... | |||
Observe the direction of the trajectory changes. Which side is this path mainly turning towards? | [
"A. Basic Straight Run",
"B. mainly turn right",
"C. mainly turn left"
] | B | single_choice | motion trend and orientation change | medium | TG_000106_OldIndustrialCity_P0000_000201_TG_C08_single_choice_motion_turn_choice_09 | OldIndustrialCity_P0000_000201 | test | TG_C08 | single_choice | TG_C08_single_choice_motion_turn_choice_09 | TG_C08_single_choice_motion_turn_choice_09_v2 | [
"A",
"B",
"C"
] | [
"Basic Straight Run",
"mainly turn right",
"mainly turn left"
] | {"A":"Basic Straight Run","B":"mainly turn right","C":"mainly turn left"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Determine the turning trend based on the change in start-end orientation angle, with evidence chain length of 2. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"turn_class":"mainly turn right","yaw_delta_deg":-165.00296},"answer_margin":165.00295671428,"gt_lineage":{"answer_margin":165.00295671428,"difficulty":"medium","difficulty_reason":"Determine the turning trend based on the change in start-end orientation angle, with evidence chain length of 2.","dis... | |||
Which candidate categories can be seen in the current image? | [
"A. lamp area",
"B. windowglass area",
"C. tree area",
"D. table area",
"E. couch area"
] | ["A","C","D"] | multi_choice | semantic category visibility | medium | TG_000024_Restaurant_P0000_000612_TG_C01_multi_choice_seg_category_multi_09 | Restaurant_P0000_000612 | test | TG_C01 | multi_choice | TG_C01_multi_choice_seg_category_multi_09 | TG_C01_multi_choice_seg_category_multi_09_v3 | [
"A",
"B",
"C",
"D",
"E"
] | [
"lamp area",
"windowglass area",
"tree area",
"table area",
"couch area"
] | {"A":"lamp area","B":"windowglass area","C":"tree area","D":"table area","E":"couch area"} | list | Set Exact Match + F1 | Exact Match / precision / recall / F1 | tartanground_gt_objective_metrics.set_scores | Compare predicted option-letter set with the GT option set; also report precision, recall and F1. | true | All visible categories must be selected from multiple candidates, with an evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"absent_distractors":["windowglass","couch"],"visible_categories":["lamp","tree","table"],"visible_pixels":{"lamp":6375,"table":11468,"tree":9402}},"answer_margin":6375.0,"gt_lineage":{"answer_margin":6375.0,"difficulty":"medium","difficulty_reason":"All visible categories must be selected from mult... | |||
which order shows the categories from largest to smallest visible area? | [
"A. floor area",
"B. ceiling area",
"C. photoframe area",
"D. sofa area"
] | ["A","B","D","C"] | ordered_list | area comparison of semantic regions | medium | TG_000056_House_P0000_000056_TG_C02_ordered_list_seg_area_order_09 | House_P0000_000056 | test | TG_C02 | ordered_list | TG_C02_ordered_list_seg_area_order_09 | TG_C02_ordered_list_seg_area_order_09_v4 | [
"A",
"B",
"C",
"D"
] | [
"floor area",
"ceiling area",
"photoframe area",
"sofa area"
] | {"A":"floor area","B":"ceiling area","C":"photoframe area","D":"sofa area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Sort the four semantic categories by visible pixel area, with evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_order_categories":["floor","ceiling","sofa","photoframe"],"area_by_category":{"ceiling":20458,"floor":36185,"photoframe":1534,"sofa":4323}},"answer_margin":2789.0,"gt_lineage":{"answer_margin":2789.0,"difficulty":"medium","difficulty_reason":"Sort the four semantic categories by visible pixe... | |||
Which candidate area has the largest visible area? | [
"A. wirespool area",
"B. tarp area",
"C. ground area",
"D. concrete column area"
] | C | single_choice | area comparison of semantic regions | medium | TG_000045_ConstructionSite_P0000_000033_TG_C02_single_choice_seg_area_largest_04 | ConstructionSite_P0000_000033 | test | TG_C02 | single_choice | TG_C02_single_choice_seg_area_largest_04 | TG_C02_single_choice_seg_area_largest_04_v5 | [
"A",
"B",
"C",
"D"
] | [
"wirespool area",
"tarp area",
"ground area",
"concrete column area"
] | {"A":"wirespool area","B":"tarp area","C":"ground area","D":"concrete column area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible pixel areas for multiple semantic categories, with evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_category":"ground","area_by_category":{"concretecolumn":28472,"ground":78700,"tarp":22244,"wirespool":2345}},"answer_margin":50228.0,"gt_lineage":{"answer_margin":50228.0,"difficulty":"medium","difficulty_reason":"Compare the visible pixel areas for multiple semantic categories, with evidenc... | |||
Please select the grid that covers the most area of the electricalbox region from the annotated area. | [
"A. Label area A",
"B. E annotated area",
"C. Annotation area F",
"D. Label region D"
] | C | grounding_choice | image region semantic localization | medium | TG_000023_NordicHarbor_P0000_000631_TG_C03_grounding_choice_grid_category_location_02 | NordicHarbor_P0000_000631 | test | TG_C03 | grounding_choice | TG_C03_grounding_choice_grid_category_location_02 | TG_C03_grounding_choice_grid_category_location_02_v1 | [
"A",
"B",
"C",
"D"
] | [
"Label area A",
"E annotated area",
"Annotation area F",
"Label region D"
] | {"A":"Label area A","B":"E annotated area","C":"Annotation area F","D":"Label region D"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Aggregate semantic category pixels into a 3x3 grid and select the region with the largest area; the evidence chain length is 2. | [
"image_<camera>/*.png",
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_region":"F","area_by_region":{"A":0,"D":75,"E":409,"F":1734},"target_category":"electricalbox"},"answer_margin":1325.0,"gt_lineage":{"answer_margin":1325.0,"difficulty":"medium","difficulty_reason":"Aggregate semantic category pixels into a 3x3 grid and select the region with the largest are... | |||
Please provide the corresponding category option letters for the specified regions. The regions to match are: H, B, I. | [
"A. road area",
"B. building area",
"C. ground pavement area"
] | {"B":"B","H":"A","I":"C"} | matching | image region semantic localization | hard | TG_000039_NordicHarbor_P0000_000473_TG_C03_matching_grid_region_category_matching_02 | NordicHarbor_P0000_000473 | test | TG_C03 | matching | TG_C03_matching_grid_region_category_matching_02 | TG_C03_matching_grid_region_category_matching_02_v1 | [
"A",
"B",
"C"
] | [
"road area",
"building area",
"ground pavement area"
] | {"A":"road area","B":"building area","C":"ground pavement area"} | mapping | DictExactStrMatchAggRecall | structured exact matching | tartanground_gt_objective_metrics.matching_scores | Parse a mapping and compare every key-value pair exactly. | true | Simultaneously determine the dominant category for three annotated regions and output the mapping, with evidence chain length of 3. | [
"image_<camera>/*.png",
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"category_options":{"A":"road area","B":"building area","C":"ground pavement area"},"region_to_category":{"B":"building","H":"road","I":"groundpavement"}},"answer_margin":7741.0,"gt_lineage":{"answer_margin":7741.0,"difficulty":"hard","difficulty_reason":"Simultaneously determine the dominant catego... | |||
Compare the typical distances by region, which cell is the nearest? | [
"A. Annotation area H",
"B. E annotated area",
"C. C labeled area",
"D. G annotated area"
] | C | grounding_choice | distance and depth ordering | medium | TG_000063_OldIndustrialCity_P0000_000252_TG_C04_grounding_choice_depth_region_nearest_08 | OldIndustrialCity_P0000_000252 | test | TG_C04 | grounding_choice | TG_C04_grounding_choice_depth_region_nearest_08 | TG_C04_grounding_choice_depth_region_nearest_08_v5 | [
"A",
"B",
"C",
"D"
] | [
"Annotation area H",
"E annotated area",
"C labeled area",
"G annotated area"
] | {"A":"Annotation area H","B":"E annotated area","C":"C labeled area","D":"G annotated area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the effective median distances of multiple annotated areas, with evidence chain length of 2. | [
"image_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"answer_region":"C","median_depth_by_region":{"C":0.50196,"E":8.02354,"G":2.00197,"H":2.00197}},"answer_margin":1.5000128746032715,"gt_lineage":{"answer_margin":1.5000128746032715,"difficulty":"medium","difficulty_reason":"Compare the effective median distances of multiple annotated areas, with evid... | |||
Which interval does the typical visible distance in area F belong to? | [
"A. 20 meters and above",
"B. less than 1 meter",
"C. 3-8 meters",
"D. 1-3 meters",
"E. 8-20 meters"
] | D | interval_choice | distance and depth ordering | simple | TG_000022_Downtown_P0000_000498_TG_C04_interval_choice_depth_region_interval_07 | Downtown_P0000_000498 | test | TG_C04 | interval_choice | TG_C04_interval_choice_depth_region_interval_07 | TG_C04_interval_choice_depth_region_interval_07_v1 | [
"A",
"B",
"C",
"D",
"E"
] | [
"20 meters and above",
"less than 1 meter",
"3-8 meters",
"1-3 meters",
"8-20 meters"
] | {"A":"20 meters and above","B":"less than 1 meter","C":"3-8 meters","D":"1-3 meters","E":"8-20 meters"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Single annotated area's valid median proximity mapping to discrete interval, evidence chain length of 1. | [
"image_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"interval":"1-3 meters","median_depth":2.00982,"region":"F","valid_fraction":1.0},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"simple","difficulty_reason":"Single annotated area's valid median proximity mapping to discrete interval, evidence chain length of 1.","distractor_str... | |||
the visible content of the signboard area is roughly at which distance range? | [
"A. 8-20 meters",
"B. 3-8 meters",
"C. 20 meters and above",
"D. 1-3 meters",
"E. less than 1 meter"
] | C | interval_choice | semantic category and combined judgment of proximity and distance | medium | TG_000076_JapaneseCity_P0000_000477_TG_C05_interval_choice_semantic_depth_interval_10 | JapaneseCity_P0000_000477 | test | TG_C05 | interval_choice | TG_C05_interval_choice_semantic_depth_interval_10 | TG_C05_interval_choice_semantic_depth_interval_10_v1 | [
"A",
"B",
"C",
"D",
"E"
] | [
"8-20 meters",
"3-8 meters",
"20 meters and above",
"1-3 meters",
"less than 1 meter"
] | {"A":"8-20 meters","B":"3-8 meters","C":"20 meters and above","D":"1-3 meters","E":"less than 1 meter"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | After applying a semantic mask to constrain the data, the near-far interval was counted, resulting in an evidence chain length of 2. | [
"seg_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"category":"signboard","interval":"20 meters and above","median_depth":32.09376,"valid_fraction":1.0},"answer_margin":1.0,"gt_lineage":{"answer_margin":1.0,"difficulty":"medium","difficulty_reason":"After applying a semantic mask to constrain the data, the near-far interval was counted, resulting in... | |||
Which category has its visible area overall more close to the observer? | [
"A. elevator area",
"B. watercontainer area",
"C. doorway area",
"D. elevatordoor area"
] | B | single_choice | semantic category and combined judgment of proximity and distance | hard | TG_000050_Hospital_P0000_000158_TG_C05_single_choice_semantic_depth_nearest_07 | Hospital_P0000_000158 | test | TG_C05 | single_choice | TG_C05_single_choice_semantic_depth_nearest_07 | TG_C05_single_choice_semantic_depth_nearest_07_v3 | [
"A",
"B",
"C",
"D"
] | [
"elevator area",
"watercontainer area",
"doorway area",
"elevatordoor area"
] | {"A":"elevator area","B":"watercontainer area","C":"doorway area","D":"elevatordoor area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Simultaneously relying on semantic mask and distance statistics for category comparison, with an evidence chain length of 3. | [
"seg_<camera>/*.png",
"depth_<camera>/*.png"
] | {"answer_evidence":{"answer_category":"watercontainer","median_depth_by_category":{"doorway":8.03916,"elevator":8.01564,"elevatordoor":8.00782,"watercontainer":2.00786}},"answer_margin":5.999962091445923,"gt_lineage":{"answer_margin":5.999962091445923,"difficulty":"hard","difficulty_reason":"Simultaneously relying on s... | |||
Please match the view labels and the candidate category options. Views to match: V, W, X, Y. | [
"A. building area",
"B. artletters area",
"C. barcounter area"
] | {"V":"A","W":"A","X":"A","Y":"A"} | matching | multi-view semantic consistency | hard | TG_000077_Restaurant_P0000_000786_TG_C06_matching_multiview_dominant_matching_05 | Restaurant_P0000_000786 | test | TG_C06 | matching | TG_C06_matching_multiview_dominant_matching_05 | TG_C06_matching_multiview_dominant_matching_05_v1 | [
"A",
"B",
"C"
] | [
"building area",
"artletters area",
"barcounter area"
] | {"A":"building area","B":"artletters area","C":"barcounter area"} | mapping | DictExactStrMatchAggRecall | structured exact matching | tartanground_gt_objective_metrics.matching_scores | Parse a mapping and compare every key-value pair exactly. | true | Simultaneously matching multiple synchronized views with the main semantic category, evidence chain length of 3. | [
"multi-camera image_<camera>",
"multi-camera seg_<camera>",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"category_options":{"A":"building area","B":"artletters area","C":"barcounter area"},"view_to_category":{"V":"building","W":"building","X":"building","Y":"building"}},"answer_margin":23333.0,"gt_lineage":{"answer_margin":23333.0,"difficulty":"hard","difficulty_reason":"Simultaneously matching multip... | |||
Which view shows the most vent area? | [
"A. Labeled View V",
"B. Labeled View W",
"C. Label view X",
"D. annotate view Y"
] | B | multi_view_choice | multi-view semantic consistency | hard | TG_000020_Downtown_P0000_001139_TG_C06_multi_view_choice_multiview_category_area_04 | Downtown_P0000_001139 | test | TG_C06 | multi_view_choice | TG_C06_multi_view_choice_multiview_category_area_04 | TG_C06_multi_view_choice_multiview_category_area_04_v1 | [
"A",
"B",
"C",
"D"
] | [
"Labeled View V",
"Labeled View W",
"Label view X",
"annotate view Y"
] | {"A":"Labeled View V","B":"Labeled View W","C":"Label view X","D":"annotate view Y"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible area of the same semantic category across four synchronized viewpoints, with evidence chain length of 3. | [
"multi-camera image_<camera>",
"multi-camera seg_<camera>",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_view":"W","area_by_view":{"V":0,"W":8418,"X":2887,"Y":692},"target_category":"vent"},"answer_margin":5531.0,"gt_lineage":{"answer_margin":5531.0,"difficulty":"hard","difficulty_reason":"Compare the visible area of the same semantic category across four synchronized viewpoints, with evidence ... | |||
Please determine the stage of path collection that point C corresponds to. | [
"A. middle segment",
"B. back segment",
"C. previous segment"
] | A | temporal_choice | trajectory order and path structure | simple | TG_000031_NordicHarbor_P0000_001420_TG_C07_temporal_choice_trajectory_stage_interval_10 | NordicHarbor_P0000_001420 | test | TG_C07 | temporal_choice | TG_C07_temporal_choice_trajectory_stage_interval_10 | TG_C07_temporal_choice_trajectory_stage_interval_10_v1 | [
"A",
"B",
"C"
] | [
"middle segment",
"back segment",
"previous segment"
] | {"A":"middle segment","B":"back segment","C":"previous segment"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Sequence position of individual path points mapped to front/middle/back stages, with evidence chain length of 1. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"frame_id":552,"normalized_rank":0.36842,"point_label":"C","stage":"middle segment"},"answer_margin":0.13157894736842107,"gt_lineage":{"answer_margin":0.13157894736842107,"difficulty":"simple","difficulty_reason":"Sequence position of individual path points mapped to front/middle/back stages, with e... | |||
Please compare the spans of each labeled section and sort them. | [
"A. Labelled Road A",
"B. Annotated segment B",
"C. labeled road segment C",
"D. Label: D"
] | ["C","B","A","D"] | ordered_list | motion trend and orientation change | hard | TG_000060_Supermarket_P0000_000640_TG_C08_ordered_list_motion_segment_speed_order_05 | Supermarket_P0000_000640 | test | TG_C08 | ordered_list | TG_C08_ordered_list_motion_segment_speed_order_05 | TG_C08_ordered_list_motion_segment_speed_order_05_v1 | [
"A",
"B",
"C",
"D"
] | [
"Labelled Road A",
"Annotated segment B",
"labeled road segment C",
"Label: D"
] | {"A":"Labelled Road A","B":"Annotated segment B","C":"labeled road segment C","D":"Label: D"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Use segment displacement length as a proxy for speed and sort at equal time intervals, with evidence chain length of 3. | [
"metadata/pose_<camera>.txt",
"frame sequence"
] | {"answer_evidence":{"answer_order_segments":["C","B","A","D"],"segment_length_by_label":{"A":7.92722,"B":8.36202,"C":9.05865,"D":3.85142}},"answer_margin":0.4348024858070305,"gt_lineage":{"answer_margin":0.4348024858070305,"difficulty":"hard","difficulty_reason":"Use segment displacement length as a proxy for speed and... | |||
Which options correspond to categories with visible pixels in the image? | [
"A. concrete floor slab area",
"B. compgrassrocksplants area",
"C. stock area",
"D. concretemixer area",
"E. constructionelevator area"
] | ["A","B","E"] | multi_choice | semantic category visibility | medium | TG_000033_ConstructionSite_P0000_000131_TG_C01_multi_choice_seg_category_multi_05 | ConstructionSite_P0000_000131 | test | TG_C01 | multi_choice | TG_C01_multi_choice_seg_category_multi_05 | TG_C01_multi_choice_seg_category_multi_05_v3 | [
"A",
"B",
"C",
"D",
"E"
] | [
"concrete floor slab area",
"compgrassrocksplants area",
"stock area",
"concretemixer area",
"constructionelevator area"
] | {"A":"concrete floor slab area","B":"compgrassrocksplants area","C":"stock area","D":"concretemixer area","E":"constructionelevator area"} | list | Set Exact Match + F1 | Exact Match / precision / recall / F1 | tartanground_gt_objective_metrics.set_scores | Compare predicted option-letter set with the GT option set; also report precision, recall and F1. | true | All visible categories must be selected from multiple candidates, with an evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"absent_distractors":["stock","concretemixer"],"visible_categories":["constructionelevator","compgrassrocksplants","concretefloorslab"],"visible_pixels":{"compgrassrocksplants":89630,"concretefloorslab":30577,"constructionelevator":10808}},"answer_margin":10808.0,"gt_lineage":{"answer_margin":10808.... | |||
Arrange the candidate categories from the largest to smallest in the frame. | [
"A. lightbulb string area",
"B. woodenbox area",
"C. petfoodtub area",
"D. wallrefrigeratordoor area"
] | ["B","D","A","C"] | ordered_list | area comparison of semantic regions | medium | TG_000075_Supermarket_P0000_000760_TG_C02_ordered_list_seg_area_order_06 | Supermarket_P0000_000760 | test | TG_C02 | ordered_list | TG_C02_ordered_list_seg_area_order_06 | TG_C02_ordered_list_seg_area_order_06_v2 | [
"A",
"B",
"C",
"D"
] | [
"lightbulb string area",
"woodenbox area",
"petfoodtub area",
"wallrefrigeratordoor area"
] | {"A":"lightbulb string area","B":"woodenbox area","C":"petfoodtub area","D":"wallrefrigeratordoor area"} | list | Exact Order Accuracy + Pairwise Accuracy | ranking | tartanground_gt_objective_metrics.ordered_scores | Full credit for exact permutation; diagnostic pairwise consistency over all ordered pairs. | true | Sort the four semantic categories by visible pixel area, with evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_order_categories":["woodenbox","wallrefrigeratordoor","lightbulbstring","petfoodtub"],"area_by_category":{"lightbulbstring":4910,"petfoodtub":1418,"wallrefrigeratordoor":5820,"woodenbox":43623}},"answer_margin":910.0,"gt_lineage":{"answer_margin":910.0,"difficulty":"medium","difficulty_reaso... | |||
Which category occupies the most pixels in the candidate semantic regions? | [
"A. building area",
"B. toothpaste region",
"C. walllight area",
"D. frozen food bagged area"
] | A | single_choice | area comparison of semantic regions | medium | TG_000065_Supermarket_P0000_000680_TG_C02_single_choice_seg_area_largest_03 | Supermarket_P0000_000680 | test | TG_C02 | single_choice | TG_C02_single_choice_seg_area_largest_03 | TG_C02_single_choice_seg_area_largest_03_v4 | [
"A",
"B",
"C",
"D"
] | [
"building area",
"toothpaste region",
"walllight area",
"frozen food bagged area"
] | {"A":"building area","B":"toothpaste region","C":"walllight area","D":"frozen food bagged area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Compare the visible pixel areas for multiple semantic categories, with evidence chain length of 2. | [
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_category":"building","area_by_category":{"building":320270,"frozenfoodbagged":5013,"toothpaste":7415,"walllight":13955}},"answer_margin":306315.0,"gt_lineage":{"answer_margin":306315.0,"difficulty":"medium","difficulty_reason":"Compare the visible pixel areas for multiple semantic categories... | |||
If only one is chosen from the A-I region, where is the plant region most concentrated? | [
"A. Label region D",
"B. E annotated area",
"C. Annotation area F",
"D. B annotated area"
] | C | grounding_choice | image region semantic localization | medium | TG_000053_ModularNeighborhood_P0000_000079_TG_C03_grounding_choice_grid_category_location_07 | ModularNeighborhood_P0000_000079 | test | TG_C03 | grounding_choice | TG_C03_grounding_choice_grid_category_location_07 | TG_C03_grounding_choice_grid_category_location_07_v1 | [
"A",
"B",
"C",
"D"
] | [
"Label region D",
"E annotated area",
"Annotation area F",
"B annotated area"
] | {"A":"Label region D","B":"E annotated area","C":"Annotation area F","D":"B annotated area"} | scalar | Accuracy | Accuracy / Acc / Overall / exact_matching | tartanground_gt_objective_metrics.score_item | Extract one capital option letter and compare with the GT answer key. | true | Aggregate semantic category pixels into a 3x3 grid and select the region with the largest area; the evidence chain length is 2. | [
"image_<camera>/*.png",
"seg_<camera>/*.png",
"seg_labels/seg_label_map.json"
] | {"answer_evidence":{"answer_region":"F","area_by_region":{"B":902,"D":2816,"E":5798,"F":9428},"target_category":"plant"},"answer_margin":3630.0,"gt_lineage":{"answer_margin":3630.0,"difficulty":"medium","difficulty_reason":"Aggregate semantic category pixels into a 3x3 grid and select the region with the largest area; ... |
End of preview. Expand in Data Studio
BenchForge TartanGround
Spatial reasoning over quadruped exploration observations.
The test split contains 10,000 English questions. The English edition preserves the original question IDs, option labels, answers, and scoring definitions. The main image and all referenced auxiliary images are embedded in the Parquet files as their original image bytes.
image: Main image supplied to the question.auxiliary_images: Ordered list of additional images supplied with the question. An empty list means the task has no additional images.question,choices, andanswer: English question and answer choices, with the original option labels in the scoring answer.data/test-00000-of-00001.jsonl: English text records with their original relative image paths for users of the existing benchmark layout.
Load the English test split using Hugging Face Datasets:
from datasets import load_dataset
dataset = load_dataset("jby20180901/BenchForge-tartanground", split="test")
Use the Parquet edition to access images directly. For the JSONL edition, use the repository’s assets/ directory or extract assets.tar at the repository root if the dataset ships with an image archive.
- Downloads last month
- 79