{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.12300", "kind": "paper", "url": "https://arxiv.org/abs/2606.12300"}, {"id": "gh:naver-ai/extremewhenbench", "kind": "repo", "url": "https://github.com/naver-ai/ExtremeWhenBench"}], "authors": ["Sukmin Seo", "Geewook Kim"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.12300", "source_url": "https://arxiv.org/abs/2606.12300"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "ExtremeWhenBench evaluates natural-language temporal grounding in hour-long videos, with 2,273 open-form queries over 194 videos. Metrics include mIoU and R@k for predicted time intervals, with strict parse-failure handling."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.12300", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.12300", "title": "ExtremeWhenBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.12300", "modality": null, "name": "ExtremeWhenBench", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "High", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "unknown", "origin_record_id": "bm_extremewhenbench_bffe4798", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "469d021b9be6f89e843ba10f767abf86ff615ef90993043f91df1196131608e8", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.12300"}, "publisher": null, "released": "2026-06-10", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.12300", "source_url": "https://arxiv.org/abs/2606.12300"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-12300", "source": "claire_radar", "source_benchmark_id": "2606.12300", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/naver-ai/ExtremeWhenBench", "githubScope": "benchmark_repo", "githubStars": 8, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2606.12300", "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", "lastSuccessfulDate": "2026-10-01", "state": "stale"}}, "status": "fresh"}, "capabilities": [], "confidence": "High", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "84a01675c6e45d379bc29c233e8795d42c78d7afbeaed6f9da63114f6a1a4248", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "ExtremeWhenBench evaluates natural-language temporal grounding in hour-long videos, with 2,273 open-form queries over 194 videos. Metrics include mIoU and R@k for predicted time intervals, with strict parse-failure handling.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "To test this, we release ExtremeWhenBench, the first open hour-scale grounding benchmark (2,273 queries over 194 videos, mean 75.7 min, max 9 hr) with an open-form query distribution."}, "familyId": "bmf_ee5095f4a257", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "existing-code-link", "linkedUrl": "https://github.com/naver-ai/ExtremeWhenBench", "url": "https://github.com/naver-ai/ExtremeWhenBench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/naver-ai/ExtremeWhenBench", "url": "https://huggingface.co/papers/2606.12300"}, {"basis": "huggingface-paper-association", "linkedUrl": "https://github.com/naver-ai/ExtremeWhenBench", "url": "https://huggingface.co/api/papers/2606.12300"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/naver-ai/ExtremeWhenBench", "url": "https://arxiv.org/abs/2606.12300"}], "scope": "benchmark_repo", "url": "https://github.com/naver-ai/ExtremeWhenBench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_extremewhenbench_bffe4798", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/naver-ai/ExtremeWhenBench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.12300", "paper": "https://arxiv.org/abs/2606.12300", "pdf": "https://arxiv.org/pdf/2606.12300", "project": null, "report": "https://arxiv.org/abs/2606.12300"}, "metrics": [], "motivation": "Temporal grounding--returning the interval $[t_s, t_e]$ for a natural-language query over a video--is the language interface to long-form video, yet has been studied on short videos; the dynamics of hour-scale natural-language grounding remain underexplored.", "name": "ExtremeWhenBench", "oneLine": "ExtremeWhenBench evaluates natural-language temporal grounding in hour-long videos, with 2,273 open-form queries over 194 videos. Metrics include mIoU and R@k for predicted time intervals, with strict parse-failure handling.", "paperTitle": "Natural-Language Temporal Grounding in Hour-Long Videos is a Search Problem: A Benchmark and Empirical Decomposition", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.12300", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "NAVER AI", "organizationType": "company-research-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/naver-ai/ExtremeWhenBench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-10", "source": {"authors": ["Sukmin Seo", "Geewook Kim"], "categories": ["cs.CV", "cs.AI"], "id": "2606.12300", "title": "Natural-Language Temporal Grounding in Hour-Long Videos is a Search Problem: A Benchmark and Empirical Decomposition", "type": "arxiv", "url": "https://arxiv.org/abs/2606.12300"}, "sourceUpdatedAt": "2026-06-10T16:35:04Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "Provides the first open hour-scale grounding benchmark to study the search problem in video understanding. Useful for developing and evaluating models for long-video retrieval and reasoning."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
