{
  "name": "NExT-QA",
  "release_date": "2021-05-01",
  "subsets": {
    "mc": {
      "language": [
        "en"
      ],
      "modalities": [
        "single_video_start"
      ],
      "task_type": "multiple_choice_vqa",
      "prompt_template": "<video>{{ question }}\n{% for k, v in options.items() %}{{ k }}. {{ v }}\n{% endfor %}Answer with the option's letter from the given choices directly.",
      "video_storage": {
        "format": "files",
        "media_root": "videos",
        "notes": "1,000 videos from VidOR dataset (out of 1,570 total in source). MC test split: 8,564 QA pairs across causal, temporal, and descriptive types."
      },
      "mapping_from_source": {
        "source": {
          "format": "huggingface",
          "url": {
            "train": "https://huggingface.co/datasets/lmms-lab/NExTQA"
          }
        },
        "id": {
          "from": "video+qid",
          "note": "Composite key MC_{video_id}_{qid}"
        },
        "question": {
          "from": "question"
        },
        "options": {
          "from": "a0-a4",
          "note": "5 choices normalized to {A,...,E} dict"
        },
        "answer": {
          "from": "answer",
          "note": "Integer index converted to letter"
        },
        "media": {
          "from": "video",
          "type": "list",
          "min_items": 1,
          "max_items": 1
        },
        "extra": {
          "video_id": {
            "from": "video"
          },
          "question_type": {
            "from": "type"
          }
        }
      },
      "prompt_template_source": {
        "origin": "official",
        "reference": "https://github.com/EvolvingLMMs-Lab/lmms-eval/blob/main/lmms_eval/tasks/nextqa/utils.py (nextqa_mc_doc_to_text — A./B./C./D./E. options + 'Answer with the option's letter from the given choices directly.')",
        "notes": "Tier 4: lmms-eval NExT-QA MC canonical evaluation prompt."
      }
    }
  }
}