{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2605.22570", "kind": "paper", "url": "https://arxiv.org/abs/2605.22570"}, {"id": "gh:zinosii/vgenst-bench", "kind": "repo", "url": "https://github.com/zinosii/VGenST-Bench"}, {"kind": "website", "url": "https://zinosii.github.io/VGenST-Bench/"}], "authors": ["Jinho Park", "Youbin Kim", "Hogun Park", "Eunbyung Park"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2605.22570", "source_url": "https://arxiv.org/abs/2605.22570"}, "categories": ["General AI", "Reasoning", "Vision & 3D"], "description": {"en": "VGenST-Bench evaluates spatio-temporal reasoning in multimodal large language models using 1,200 procedurally generated videos with controlled scene composition, camera trajectory, and reasoning targets. It covers 12 reasoning tasks across three spatial scales and 12 QA types across three reasoning levels, with multiple choice and open-ended question variants."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2605.22570", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2605.22570", "title": "VGenST-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2605.22570", "modality": null, "name": "VGenST-Bench", "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_vgenst-bench_f8fb9973", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "a892f33fd5551eb22370e78056e4ea98df436c9debc25e441a2a945084618440", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2605.22570"}, "publisher": null, "released": "2026-05-21", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2605.22570", "source_url": "https://arxiv.org/abs/2605.22570"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2605-22570", "source": "claire_radar", "source_benchmark_id": "2605.22570", "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/zinosii/VGenST-Bench", "githubScope": "benchmark_repo", "githubStars": 14, "hfDailySubmittedAt": "2026-05-25T00:00:00.000Z", "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 22, "hfPaperUrl": "https://huggingface.co/papers/2605.22570", "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": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "capabilities": ["Reasoning"], "confidence": "High", "construction": "Transform Existing", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "da5d35794b1ed6f3b0129bc5f428a0f94f85a866a439b03f27a0decc4c7653cf", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "VGenST-Bench evaluates spatio-temporal reasoning in multimodal large language models using 1,200 procedurally generated videos with controlled scene composition, camera trajectory, and reasoning targets. It covers 12 reasoning tasks across three spatial scales and 12 QA types across three reasoning levels, with multiple choice and open-ended question variants.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["coined title prefix ending in Bench or Benchmark", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "In this paper, we introduce VGenST-Bench, a video benchmark that employs generative models to actively synthesize highly controlled and diverse evaluation scenarios."}, "familyId": "bmf_3236a25f97a3", "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/zinosii/VGenST-Bench", "url": "https://github.com/zinosii/VGenST-Bench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/zinosii/VGenST-Bench", "url": "https://huggingface.co/papers/2605.22570"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/zinosii/VGenST-Bench", "url": "https://zinosii.github.io/VGenST-Bench/"}], "scope": "benchmark_repo", "url": "https://github.com/zinosii/VGenST-Bench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_vgenst-bench_f8fb9973", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/zinosii/VGenST-Bench", "data": null, "hfPaper": "https://huggingface.co/papers/2605.22570", "paper": "https://arxiv.org/abs/2605.22570", "pdf": "https://arxiv.org/pdf/2605.22570", "project": "https://zinosii.github.io/VGenST-Bench/", "report": "https://arxiv.org/abs/2605.22570"}, "metrics": [], "motivation": "Spatio-temporal reasoning is a core capability for Multimodal Large Language Models (MLLMs) operating in the real world.", "name": "VGenST-Bench", "oneLine": "VGenST-Bench evaluates spatio-temporal reasoning in multimodal large language models using 1,200 procedurally generated videos with controlled scene composition, camera trajectory, and reasoning targets. It covers 12 reasoning tasks across three spatial scales and 12 QA types across three reasoning levels, with multiple choice and open-ended question variants.", "paperTitle": "VGenST-Bench: A Benchmark for Spatio-Temporal Reasoning via Active Video Synthesis", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2605.22570", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Sungkyunkwan University", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://zinosii.github.io/VGenST-Bench/"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-05-21", "source": {"authors": ["Jinho Park", "Youbin Kim", "Hogun Park", "Eunbyung Park"], "categories": ["cs.CV", "cs.AI"], "id": "2605.22570", "title": "VGenST-Bench: A Benchmark for Spatio-Temporal Reasoning via Active Video Synthesis", "type": "arxiv", "url": "https://arxiv.org/abs/2605.22570"}, "sourceUpdatedAt": "2026-05-21T14:48:35Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "Existing video reasoning benchmarks rely on static or passively curated content, limiting fine-grained diagnosis. VGenST-Bench uses active synthesis to enable controlled and diverse evaluation of fine-grained spatio-temporal reasoning, supporting model comparison and targeted improvement in MLLMs."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
