{
  "schema_version": "1.0",
  "id": "archive:https://arxiv.org/abs/2607.11862v1",
  "slug": "2607-11862v1-18as4nc",
  "url": "https://feed7.dev/p/2607-11862v1-18as4nc",
  "title": "Evidence-Backed Video Question Answering",
  "why_included": "E-VQA requires video answers to include tracked pixel-level evidence, revealing when good QA scores hide weak perception and supplying grounded training data.",
  "summary": "**E-VQA** requires a semantic answer plus temporal segments and dense tracked segmentation masks. Its human-verified ST-Evidence benchmark finds that QA accuracy can diverge from visual grounding, a gap the authors say scaling alone does not close.",
  "practical_implication": "For video agents, evaluate whether the cited object persists through motion, occlusion, and deformation rather than scoring text alone. The accompanying **160k-scale ST-Evidence-Instruct** dataset provides training examples that tie reasoning to visible evidence.",
  "agent_context": "**E-VQA** requires a semantic answer plus temporal segments and dense tracked segmentation masks. Its human-verified ST-Evidence benchmark finds that QA accuracy can diverge from visual grounding, a gap the authors say scaling alone does not close.\n\nFor video agents, evaluate whether the cited object persists through motion, occlusion, and deformation rather than scoring text alone. The accompanying **160k-scale ST-Evidence-Instruct** dataset provides training examples that tie reasoning to visible evidence.\n\nOn a **7B model**, fine-tuning beats a size-matched UniPixel baseline by **+27.2 t-mean and +13.8 J&F**. Those gains are specific to the reported setup and grounding metrics; the material does not establish equivalent gains for downstream video-agent tasks.",
  "source": {
    "name": "arXiv",
    "url": "https://arxiv.org/abs/2607.11862v1",
    "published_at": null
  },
  "source_class": "blog_post",
  "content_type": "Paper",
  "layer": "benchmark",
  "domains": [
    "video"
  ],
  "topics": [
    "generative-media",
    "agent-evals",
    "benchmark-integrity"
  ],
  "verification": {
    "status": "needs_review",
    "label": "Needs Review",
    "method": "unverified",
    "verified_at": null
  },
  "uncertainty": [
    "On a **7B model**, fine-tuning beats a size-matched UniPixel baseline by **+27.2 t-mean and +13.8 J&F**. Those gains are specific to the reported setup and grounding metrics; the material does not establish equivalent gains for downstream video-agent tasks."
  ],
  "lifecycle": "Current",
  "published_at": null,
  "modified_at": null,
  "supersedes": [],
  "expires_at": null,
  "formats": {
    "html": "https://feed7.dev/p/2607-11862v1-18as4nc",
    "json": "https://feed7.dev/p/2607-11862v1-18as4nc.json",
    "markdown": "https://feed7.dev/p/2607-11862v1-18as4nc.md"
  }
}