{"record": {"aliases": ["Evidence-Backed Video Question Answering"], "artifacts": [{"id": "arxiv:2607.11862", "kind": "paper", "url": "https://arxiv.org/abs/2607.11862"}, {"id": "gh:salesforceairesearch/evqa", "kind": "repo", "url": "https://github.com/SalesforceAIResearch/EVQA"}], "authors": ["Shijie Wang", "Honglu Zhou", "Ziyang Wang", "Ran Xu", "Caiming Xiong", "Silvio Savarese", "Chen Sun", "Juan Carlos Niebles"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.11862", "source_url": "https://arxiv.org/abs/2607.11862"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "ST-Evidence evaluates video question answering that must return both semantic answers and dense spatio-temporal visual evidence."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.11862", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.11862", "title": "ST-Evidence"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.11862", "modality": null, "name": "ST-Evidence", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-09-25T04:40:22Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_evidence-backed-video-question-answering_ce2538f7", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "06cf41dd704f92c2935dcf9135224c7aa0e4d6fcda41628d40adb864480fac3e", "relation": "introduces", "review_model": "codex-cli-default", "review_state": "ai-reviewed", "reviewed_at": "2026-09-24T17:27:47.088316Z", "source_url": "https://arxiv.org/abs/2607.11862"}, "publisher": null, "released": "2026-07-13", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.11862", "source_url": "https://arxiv.org/abs/2607.11862"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-11862", "source": "claire_radar", "source_benchmark_id": "2607.11862", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["Evidence-Backed Video Question Answering"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The combination of video reasoning, dense grounding, human verification, and released resources is likely to draw near-term interest.", "score": 4}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-24T17:27:47.088316Z", "inputHash": "5a464a9c009f67259b870ee1dc7127ef01f89dc15442c70e2eeffc6a0a3a86d8", "model": "codex-cli-default"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "ST-Evidence, the first human-verified benchmark for both discriminative and generative pixel-level grounding", "canonicalNameSource": "abstract", "decisionReason": "The abstract formally names the benchmark, specifies its joint answer-and-grounding task, reports named evaluation measures, and supplies a public code-and-data path.", "evaluationEvidence": {"reuse": "Code and data are available at https://github.com/SalesforceAIResearch/EVQA.", "scoring": "+27.2 t-mean and +13.8 J&F on a 7B model", "task": "requiring models to jointly output a semantic answer and precise spatio-temporal evidence: temporal segments and dense, tracked object segmentation masklets"}, "model": "codex-cli-default", "reviewedAt": "2026-09-24T17:27:47.088316Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "ST-Evidence evaluates video question answering that must return both semantic answers and dense spatio-temporal visual evidence.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "To support this, we introduce ST-Evidence, the first human-verified benchmark for both discriminative and generative pixel-level grounding."}, "familyId": "bmf_953b54c9e131", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_evidence-backed-video-question-answering_ce2538f7", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/SalesforceAIResearch/EVQA", "data": null, "paper": "https://arxiv.org/abs/2607.11862", "pdf": "https://arxiv.org/pdf/2607.11862", "project": null, "report": "https://arxiv.org/abs/2607.11862"}, "metrics": [], "motivation": "Current Video Large Language Models (Video LLMs) excel in question answering (QA) but largely operate as black boxes, providing textual answers without verifiable visual grounding.", "name": "ST-Evidence", "oneLine": "ST-Evidence evaluates video question answering that must return both semantic answers and dense spatio-temporal visual evidence.", "paperTitle": "Evidence-Backed Video Question Answering", "primaryDomain": "General AI", "publication": {"evidence": "ECCV 2026", "evidenceLevel": "strong-author-metadata", "evidenceUrl": "https://arxiv.org/abs/2607.11862", "source": "arxiv-journal-reference", "status": "publication_reported", "venue": "ECCV 2026", "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [{"evidence": [{"level": "strong-author-metadata", "observedAt": "2026-08-19T11:05:13.395391Z", "rawValue": "ECCV 2026", "sourceType": "arxiv-journal-reference", "sourceUrl": "https://arxiv.org/abs/2607.11862"}], "publicationStatus": "published", "venueName": "ECCV 2026"}], "publishers": [{"name": "Salesforce AI Research", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/SalesforceAIResearch/EVQA"}], "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-07-13", "reviewLifecycle": {"state": "pending"}, "source": {"authors": ["Shijie Wang", "Honglu Zhou", "Ziyang Wang", "Ran Xu", "Caiming Xiong", "Silvio Savarese", "Chen Sun", "Juan Carlos Niebles"], "categories": ["cs.CV", "cs.AI"], "id": "2607.11862", "title": "Evidence-Backed Video Question Answering", "type": "arxiv", "url": "https://arxiv.org/abs/2607.11862"}, "sourceUpdatedAt": "2026-07-13T17:49:10Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "It exposes whether strong question-answering results correspond to verifiable perception and grounding in video."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
