{"record": {"aliases": ["Post-Training VLMs for Video Mistake Detection"], "artifacts": [{"id": "arxiv:2608.28406", "kind": "paper", "url": "https://arxiv.org/abs/2608.28406"}, {"id": "gh:fedespu/mstk", "kind": "repo", "url": "https://github.com/FedeSpu/mstk"}], "authors": ["Federico Spurio", "Olga Zatsarynna", "Lars Doorenbos", "Emad Bahrami", "Gianpiero Francesca", "Juergen Gall"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.28406", "source_url": "https://arxiv.org/abs/2608.28406"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "Evaluates mistake detection in videos via question answering, testing whether methods can discern if a step was executed correctly against its description for seen and unseen actions."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.28406", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.28406", "title": "MD-VQA"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of Bonn", "Lamarr Institute for Machine Learning and Artificial Intelligence", "Toyota Motor Corporation (Belgium)"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2608.28406", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2608.28406", "source_url": "https://openalex.org/W7204757332"}, "key": "claire-radar:2608.28406", "modality": null, "name": "MD-VQA", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "false", "origin_record_id": "bm_post-training-vlms-for-video-mistake-detec_58e7640e", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "c7855c96f8e6bd6505e929ad679d9cc3cbc66a81782f089fcc1a5ec3703da913", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-name-audit-deferred", "reviewed_at": "2026-09-05T08:28:38.611381Z", "source_url": "https://arxiv.org/abs/2608.28406"}, "publisher": null, "released": "2026-08-28", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.28406", "source_url": "https://arxiv.org/abs/2608.28406"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-28406", "source": "claire_radar", "source_benchmark_id": "2608.28406", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["Post-Training VLMs for Video Mistake Detection"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/FedeSpu/mstk", "githubScope": "benchmark_repo", "githubStars": 4, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2608.28406", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "state": "unavailable"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Low", "horizon": "7d", "reason": "The benchmark addresses a general mistake detection concept across seen and unseen actions, but public reuse evidence is unverified, limiting forecast confidence.", "score": 55}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-05T08:28:38.611381Z", "inputHash": "820788b43abe4333560c81b86a3caaed0c70f5ce44c2649c0587985ce0fa29c1", "model": "deepseek-v4-pro"}, "curation": {"canonicalNameEvidence": "we introduce the Mistake Detection Video Question Answering (MD-VQA) protocol and accompanying benchmark.", "canonicalNameSource": "abstract", "decisionReason": "Artifact link is unverified and no independent confirmation of a stable public benchmark; deferring pending clearer release evidence.", "model": "deepseek-v4-pro", "reviewedAt": "2026-09-05T08:28:38.611381Z", "state": "ai-name-audit-deferred"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Evaluates mistake detection in videos via question answering, testing whether methods can discern if a step was executed correctly against its description for seen and unseen actions.", "displayEligible": false, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "score_submission", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "To reflect this, we introduce the Mistake Detection Video Question Answering (MD-VQA) protocol and accompanying benchmark."}, "familyId": "bmf_5cfb2a67196a", "firstSeenAt": "2026-08-31", "heat": null, "id": "bm_post-training-vlms-for-video-mistake-detec_58e7640e", "indexedAt": "2026-08-31T01:03:30.163531Z", "industrySectors": [], "links": {"code": "https://github.com/FedeSpu/mstk", "data": null, "paper": "https://arxiv.org/abs/2608.28406", "pdf": "https://arxiv.org/pdf/2608.28406", "project": null, "report": "https://arxiv.org/abs/2608.28406"}, "metrics": [], "motivation": "Human mistakes are inevitable when following instructions, yet they can lead to severe consequences.", "name": "MD-VQA", "oneLine": "Evaluates mistake detection in videos via question answering, testing whether methods can discern if a step was executed correctly against its description for seen and unseen actions.", "paperTitle": "Post-Training VLMs for Video Mistake Detection", "primaryDomain": "General AI", "publication": {"evidence": "Accepted at BMVC 2026", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2608.28406", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "BMVC 2026", "verifiedAt": "2026-08-31T01:03:30.163531Z"}, "publications": [], "ranking": {"90d": {"confidence": "Low", "coverage": 0.55, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.21483102985913266, "value": 4}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 197, "score": 34}, "method": "attention-ranking-v13", "rank": 197, "score": 34}}, "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-08-28", "source": {"authors": ["Federico Spurio", "Olga Zatsarynna", "Lars Doorenbos", "Emad Bahrami", "Gianpiero Francesca", "Juergen Gall"], "categories": ["cs.CV", "cs.LG"], "id": "2608.28406", "title": "Post-Training VLMs for Video Mistake Detection", "type": "arxiv", "url": "https://arxiv.org/abs/2608.28406"}, "sourceUpdatedAt": "2026-08-28T14:56:58Z", "topics": ["cs.CV"], "venueAttempts": [{"decisionRaw": "Accepted at BMVC 2026", "evidence": [{"level": "author-claim", "observedAt": "2026-08-31T01:03:30.163531Z", "rawValue": "Accepted at BMVC 2026", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2608.28406"}], "reviewStatus": "accepted", "venueName": "BMVC 2026"}], "whyItMatters": "Provides a protocol to assess generalization of mistake detection beyond closed-set protocols, addressing a key limitation for broader applicability."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
