| { | |
| "name": "WorldMedQA-V", | |
| "release_date": "2024-10-15", | |
| "subsets": { | |
| "default": { | |
| "language": [ | |
| "en" | |
| ], | |
| "modalities": [ | |
| "single_image_start" | |
| ], | |
| "task_type": "multiple_choice_qa", | |
| "score_type": "llm_extract", | |
| "score_protocol": { | |
| "reference": "official@github.com/WorldMedQA/V (README: 'All code for handling and evaluating the dataset is available in the VLMEvalKit repository') -> vlmevalkit@vlmeval/dataset/image_mcq.py:114 WorldMedQA-V is registered in ImageMCQDataset.DATASET_URL, so evaluate() uses the base ImageMCQDataset.evaluate -> mcq_vanilla_eval (rule can_infer first, GPT judge extraction fallback). Confirmed in clone (no WorldMedQA-V subclass; base MCQ path).", | |
| "note": "Multilingual medical MCQ (A-D, options in extra fields A/B/C/D; multiple language rows per image). Authors delegate evaluation to VLMEvalKit's standard MCQ flow (LLM choice extractor). Headline is overall accuracy across languages/local-vs-English." | |
| }, | |
| "prompt_template": "<image>{{ question }}\nA. {{ options.A }}\nB. {{ options.B }}\nC. {{ options.C }}\nD. {{ options.D }}\nAnswer with the option's letter from the given choices directly.", | |
| "prompt_template_source": { | |
| "origin": "official", | |
| "reference": "https://huggingface.co/datasets/WorldMedQA/V", | |
| "notes": "Tier 1: WorldMedQA-V multilingual medical MCQ with A-D options and correct_option letter; standard medical-VQA prompt." | |
| }, | |
| "mapping_from_source": { | |
| "media": { | |
| "from": "image", | |
| "type": "list", | |
| "min_items": 1, | |
| "max_items": 1 | |
| }, | |
| "id": { | |
| "from": "index" | |
| }, | |
| "question": { | |
| "from": "question" | |
| }, | |
| "answer": { | |
| "from": "correct_option", | |
| "optional": true | |
| }, | |
| "extra": { | |
| "A": { | |
| "from": "A" | |
| }, | |
| "B": { | |
| "from": "B" | |
| }, | |
| "C": { | |
| "from": "C" | |
| }, | |
| "D": { | |
| "from": "D" | |
| }, | |
| "split_col": { | |
| "from": "split" | |
| } | |
| }, | |
| "source": { | |
| "format": "huggingface", | |
| "url": { | |
| "train": "https://huggingface.co/datasets/WorldMedQA/V" | |
| } | |
| } | |
| } | |
| } | |
| } | |
| } | |