File size: 2,052 Bytes
5433bb1 81ffd67 5433bb1 81ffd67 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 | {
"name": "POPE",
"release_date": "2026-05-15",
"subsets": {
"main": {
"language": [
"en"
],
"modalities": [
"single_image_start"
],
"task_type": "yes_no_qa",
"score_pipeline": [
"exact-match",
"rule-match"
],
"score_protocol": {
"reference": "lmms-eval@lmms_eval/tasks/pope/utils.py:17-29 — pred.lower().strip() exact-compared to gt yes/no, no LLM; aggregations utils.py:32-90 compute accuracy, precision, recall, F1, yes-ratio.",
"note": "Official headline is accuracy+precision+recall+F1 reported per category (adversarial/popular/random, carried in extra.category); per-sample accuracy reproduces only accuracy — F1/precision/recall must be recomputed at aggregation (VLMEvalKit yorn.py:148-187 POPE_rating explodes per category). VLMEvalKit variant adds GPT yes/no extraction fallback (yorn.py:254-272 YOrN_Extraction + YOrN_auxeval); lmms-eval/official use pure rules — rule chosen per official protocol."
},
"prompt_template": "<image>{{ question }}",
"mapping_from_source": {
"media": {
"from": "image",
"type": "list",
"min_items": 1,
"max_items": 1
},
"id": {
"from": "id"
},
"question": {
"from": "question"
},
"answer": {
"from": "answer",
"optional": true
},
"extra": {
"category": {
"from": "category"
}
},
"source": {
"format": "json",
"url": {
"test": "https://huggingface.co/datasets/lmms-lab/POPE"
}
}
},
"prompt_template_source": {
"origin": "official",
"reference": "https://github.com/open-compass/VLMEvalKit/blob/main/vlmeval/dataset/image_yorn.py (ImageYORNDataset.build_prompt — bare yes/no question)",
"notes": "Tier 3: VLMEvalKit POPE ImageYORNDataset.build_prompt: bare yes/no question."
}
}
}
} |