File size: 2,412 Bytes
cb7ef25 19a6bbc cb7ef25 19a6bbc | 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 | {
"name": "PathVQA",
"release_date": "2003-03-30",
"subsets": {
"main": {
"language": [
"en"
],
"modalities": [
"single_image_start"
],
"task_type": "yes_no_qa",
"score_pipeline": [
"exact-match",
"rule-match",
"llm-judge"
],
"score_protocol": {
"reference": "VLMEvalKit@vlmeval/dataset/image_shortqa.py:144-160 (PathVQA_VAL/PathVQA_TEST -> ImageShortQADataset.evaluate, lines 87-142) verified: grades short answers with an LLM judge (build_judge default gpt-4o-mini via Comprehensive_auxeval / ShortQA_auxeval line 15), with exact_matching fallback when model=='exact_matching' or the API is down. Returns binary correctness (report_acc).",
"note": "Official/standard med-VQA reports closed-ended (yes/no) accuracy plus open-ended token-recall/F1; VLMEvalKit ships an LLM-judge correctness grader. Modeled here as rule_llm_judge: deterministic exact match resolves closed yes/no rows, LLM judge grades open answers. Token-F1 open metric is not reproduced per-sample."
},
"prompt_template": "<image>{{ question }}\nPlease directly provide a short answer to the question. ",
"mapping_from_source": {
"media": {
"from": "image",
"type": "list",
"min_items": 1,
"max_items": 1
},
"id": {
"from": "-"
},
"question": {
"from": "question"
},
"answer": {
"from": "answer",
"optional": true
},
"source": {
"format": "huggingface",
"url": {
"train": "https://huggingface.co/datasets/flaviagiammarino/path-vqa",
"validation": "https://huggingface.co/datasets/flaviagiammarino/path-vqa",
"test": "https://huggingface.co/datasets/flaviagiammarino/path-vqa"
}
}
},
"prompt_template_source": {
"origin": "official",
"reference": "VLMEvalKit vlmeval/dataset/image_shortqa.py#L80-L84 (ImageShortQADataset.build_prompt, byte-verified in clone; PathVQA registered as ShortQA)",
"notes": "Tier 3: VLMEvalKit ShortQA trailer (with its official trailing space). Fixed 2026-07-07 (previous LLaVA-1.5-style trailer matched no source; LLaVA-Med feeds the raw question but is a model harness, not a benchmark prompt spec)."
}
}
}
} |