| { | |
| "name": "MMVP", | |
| "release_date": "2024-01-01", | |
| "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:249-324 (ImageMCQDataset.evaluate_heuristic) -> vlmevalkit@vlmeval/dataset/utils/multiple_choice.py:475-508 mcq_vanilla_eval: can_infer rule prefetch, then GPT extractor (extract_answer_from_item @vlmeval/dataset/utils/multiple_choice.py:359-407) maps verbose output to a letter, rule-compare vs GT; MMVP registered @vlmeval/dataset/image_mcq.py:121; headline via report_acc_MMVP @vlmeval/dataset/utils/multiple_choice.py:613-622: pairs of consecutive rows (2i, 2i+1) — 'Overall' counts a pair only if BOTH hits are 1", | |
| "note": "Official headline 'Overall' is PAIR accuracy (a CLIP-blind pair scores only if both questions are correct); a per-sample scorer reports per-question accuracy, systematically higher. Pair = consecutive ids (mm-eval ids 1..300, pair k = ids 2k-1 & 2k). GT stored as lowercase parenthesized letters '(a)'/'(b)' with options in options_text ('(a) Open (b) Closed'); matcher must normalize case/parens." | |
| }, | |
| "prompt_template": "<image>{{ question }} {{ options.options_text }}", | |
| "prompt_template_source": { | |
| "origin": "official", | |
| "reference": "https://github.com/tsb0601/MMVP/blob/main/scripts/evaluate_mllm.py (cur_prompt = row['Question'] + ' ' + row['Options'])", | |
| "notes": "Tier 1: MMVP official evaluation prompt per tsb0601/MMVP/scripts/evaluate_mllm.py line 59: cur_prompt = row['Question'] + ' ' + row['Options']. No trailer is appended by the official script. Options field uses '(a) Open (b) Closed' inline format from the source CSV/HF parquet." | |
| }, | |
| "mapping_from_source": { | |
| "media": { | |
| "from": "image", | |
| "type": "list", | |
| "min_items": 1, | |
| "max_items": 1 | |
| }, | |
| "id": { | |
| "from": "Index" | |
| }, | |
| "question": { | |
| "from": "Question" | |
| }, | |
| "answer": { | |
| "from": "Correct Answer", | |
| "optional": true | |
| }, | |
| "extra": { | |
| "options_text": { | |
| "from": "Options", | |
| "note": "Pre-rendered options string in '(a) opt1 (b) opt2' format from MMVP source." | |
| } | |
| }, | |
| "source": { | |
| "format": "huggingface", | |
| "url": { | |
| "test": "https://huggingface.co/datasets/lmms-lab-eval/MMVP" | |
| } | |
| } | |
| } | |
| } | |
| } | |
| } |