CVQA / metadata.json
ZTWHHH's picture
metadata: migrate score_type -> score_pipeline (atomic stage contract; see mm-eval scorer docs/en/SCORING.md)
d944dd4 verified
Raw
History Blame Contribute Delete
2.98 kB
{
"name": "CVQA",
"release_date": "2024-06-08",
"subsets": {
"main": {
"language": [
"en"
],
"modalities": [
"single_image_start"
],
"task_type": "multiple_choice_qa",
"score_pipeline": [
"rule-match",
"llm-match"
],
"score_protocol": {
"reference": "vlmevalkit@vlmeval/dataset/image_mcq.py:3065 (class CVQA(ImageMCQDataset)) + image_mcq.py:255-330 evaluate_heuristic -> utils/multiple_choice.py:475 mcq_vanilla_eval — exact letter matching first, GPT choice-extraction fallback, then rule compare; CVQA has no public official eval code (leaderboard), VLMEvalKit is the highest available tier.",
"note": "mm-eval ships the local-language question variant (VLMEvalKit CVQA_LOC equivalent; translated_question/translated_options kept in extra). Official paper reports local-language and English-translated prompts as separate tracks."
},
"prompt_template": "<image>Question: {{ question }}\nOptions:\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Select the best answer to the above multiple-choice question based on the image. Respond with only the letter of the correct option (A, B, C, or D).\nThe best answer is: ",
"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
},
"options": {
"from": "options",
"optional": true,
"note": "list source values are normalized to {A,B,...} dict"
},
"extra": {
"translated_question": {
"from": "translated_question"
},
"translated_options": {
"from": "translated_options"
},
"subset": {
"from": "subset"
},
"category": {
"from": "category"
},
"image_type": {
"from": "image_type"
}
},
"source": {
"format": "huggingface",
"url": {
"test": "https://huggingface.co/datasets/afaji/cvqa"
}
}
},
"prompt_template_source": {
"origin": "official",
"reference": "VLMEvalKit vlmeval/dataset/image_mcq.py#L3088-L3123 (CVQA class build_prompt, byte-verified in clone)",
"notes": "Tier 3: deployed the byte-verifiable VLMEvalKit CVQA prompt. The authors' paper (arXiv:2406.05967 §3) publishes a T1 prompt ('Question: {q} Options: {options} Short Answer:') whose options byte-format is under-specified in the paper text — flagged as an open question rather than deploying an invented rendering. Fixed 2026-07-07 (previous template cited a nonexistent lmms-eval task)."
}
}
}
}