{
  "aliases": [],
  "authors": [],
  "details": {
    "dataset_profile": {
      "counts": [
        {
          "basis": "reported",
          "evidence_url": "https://raw.githubusercontent.com/google-research-datasets/screen_qa/HEAD/README.md",
          "population": "original ScreenQA train split",
          "qualifier": "exact",
          "unit": "question-answer pairs",
          "value": 68951
        },
        {
          "basis": "reported",
          "evidence_url": "https://raw.githubusercontent.com/google-research-datasets/screen_qa/HEAD/README.md",
          "population": "original ScreenQA validation split",
          "qualifier": "exact",
          "unit": "question-answer pairs",
          "value": 8614
        },
        {
          "basis": "reported",
          "evidence_url": "https://raw.githubusercontent.com/google-research-datasets/screen_qa/HEAD/README.md",
          "population": "original ScreenQA test split",
          "qualifier": "exact",
          "unit": "question-answer pairs",
          "value": 8419
        }
      ],
      "evidence_urls": [
        "https://raw.githubusercontent.com/google-research-datasets/screen_qa/HEAD/README.md"
      ],
      "formats": [
        "JSON"
      ],
      "modalities": [
        "questions",
        "answers",
        "screen references",
        "bounding boxes"
      ],
      "notes": "All questions for a given screenshot share one split. ScreenQA Short reuses the same questions/screens; ComplexQA is a separate extension. Do not add these variants as independent original examples.",
      "splits": [
        "train",
        "validation",
        "test"
      ],
      "supervision": [
        "screen question answering",
        "answer grounding"
      ]
    }
  },
  "environments": [
    "android"
  ],
  "evidence": [
    {
      "method": "editorial_primary_review",
      "retrieved_at": "2026-10-16T06:36:01Z",
      "role": "primary",
      "sha256": "7d1bfe7d6eddd5d3fc4c33d4814de3cabf7a6b6fd3d3f85ba4c2b33036bf8867",
      "supports": [
        "name",
        "kind",
        "summary",
        "environments",
        "topics",
        "scope",
        "relations",
        "details.dataset_profile"
      ],
      "url": "https://raw.githubusercontent.com/google-research-datasets/screen_qa/HEAD/README.md"
    }
  ],
  "id": "github:google-research-datasets/screen_qa",
  "identifiers": {},
  "kind": "dataset",
  "last_verified_at": "2026-10-16T06:36:01Z",
  "licenses": {
    "code": null,
    "data": null,
    "document": null,
    "weights": null
  },
  "links": [
    {
      "rel": "primary",
      "url": "https://github.com/google-research-datasets/screen_qa"
    }
  ],
  "name": "ScreenQA",
  "notes": "",
  "published_at": null,
  "relations": [
    {
      "evidence_url": "https://raw.githubusercontent.com/google-research-datasets/screen_qa/HEAD/README.md",
      "target_id": "arxiv:2209.08199",
      "type": "documented_by"
    }
  ],
  "review_status": "curated",
  "scope": "core",
  "summary": "About 86,000 human-written question-answer pairs over 35,000 Rico mobile screenshots, plus ScreenQA Short and ComplexQA variants, for benchmarking screen content understanding.",
  "topics": [
    "screen-understanding"
  ],
  "updated_at": null,
  "url": "https://github.com/google-research-datasets/screen_qa"
}