{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.02991", "kind": "paper", "url": "https://arxiv.org/abs/2607.02991"}, {"id": "gh:fawnliu/guideme", "kind": "repo", "url": "https://github.com/fawnliu/GuideMe"}, {"kind": "website", "url": "https://fawnliu.github.io/project/guideme"}], "authors": ["Fang Liu", "Jinpeng Chen", "Ke Xu", "Yuhao Liu", "Huankang Guan", "Xudong Lu", "Bo Yang", "Gerhard Hancke", "Rui Liu", "Rynson W. H. Lau"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.02991", "source_url": "https://arxiv.org/abs/2607.02991"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "GuideMe is a benchmark for evaluating multimodal large language models on streaming video task guidance. It includes 2,458 videos (223.7 hours) with 47,775 interaction samples covering next-step instructions, completion feedback, error detection, and corrective guidance. Assessment uses temporal-semantic matching, behavioral classification, and LLM-as-a-Judge."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.02991", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.02991", "title": "GuideMe"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.02991", "modality": null, "name": "GuideMe", "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_guideme_feb34852", "origin_source": "arxiv", "recognition_confidence": "0.95", "record_sha256": "61da2bbd4e263b3726d76935cec812c00d42edbcb7f7568a5bc6dfae0ff21499", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2607.02991"}, "publisher": null, "released": "2026-07-03", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.02991", "source_url": "https://arxiv.org/abs/2607.02991"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-02991", "source": "claire_radar", "source_benchmark_id": "2607.02991", "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/fawnliu/GuideMe", "githubScope": "benchmark_repo", "githubStars": 4, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 1, "hfPaperUrl": "https://huggingface.co/papers/2607.02991", "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": "2026-10-04", "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": "c6bf1e418be906dd2d54481190c0162801ab27f343cd9276d951faa6cda8e553", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "GuideMe is a benchmark for evaluating multimodal large language models on streaming video task guidance. It includes 2,458 videos (223.7 hours) with 47,775 interaction samples covering next-step instructions, completion feedback, error detection, and corrective guidance. Assessment uses temporal-semantic matching, behavioral classification, and LLM-as-a-Judge.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact coined title identity tied to benchmark evidence", "evaluation protocol evidence", "public artifact URL"], "snippet": "In this paper, we construct GuideMe, the first multi-domain benchmark for streaming video that supports training and evaluation of MLLMs for closed-loop interactive task guidance."}, "familyId": "bmf_f4f8f3df85f0", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 1, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "linked-from-project", "linkedUrl": "https://github.com/fawnliu/GuideMe", "url": "https://fawnliu.github.io/project/guideme"}], "scope": "benchmark_repo", "url": "https://github.com/fawnliu/GuideMe", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_guideme_feb34852", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/fawnliu/GuideMe", "data": null, "hfPaper": "https://huggingface.co/papers/2607.02991", "paper": "https://arxiv.org/abs/2607.02991", "pdf": "https://arxiv.org/pdf/2607.02991", "project": "https://fawnliu.github.io/project/guideme", "report": "https://arxiv.org/abs/2607.02991"}, "metrics": [], "motivation": "While multimodal Large Language Models (MLLMs) excel at offline video understanding, an interesting question of how far they are from serving as a real-time procedural coach remains unknown.", "name": "GuideMe", "oneLine": "GuideMe is a benchmark for evaluating multimodal large language models on streaming video task guidance. It includes 2,458 videos (223.7 hours) with 47,775 interaction samples covering next-step instructions, completion feedback, error detection, and corrective guidance. Assessment uses temporal-semantic matching, behavioral classification, and LLM-as-a-Judge.", "paperTitle": "GuideMe: Multi-Domain Task Guidance and Intervention in Streaming Video", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.02991", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "GuideMe Project", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://fawnliu.github.io/project/guideme"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 0.95, "relation": "introduces", "releasedAt": "2026-07-03", "source": {"authors": ["Fang Liu", "Jinpeng Chen", "Ke Xu", "Yuhao Liu", "Huankang Guan", "Xudong Lu", "Bo Yang", "Gerhard Hancke", "Rui Liu", "Rynson W.H. Lau"], "categories": ["cs.CV"], "id": "2607.02991", "title": "GuideMe: Multi-Domain Task Guidance and Intervention in Streaming Video", "type": "arxiv", "url": "https://arxiv.org/abs/2607.02991"}, "sourceUpdatedAt": "2026-07-03T06:00:04Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "Existing multimodal models lack closed-loop interactive coaching ability; GuideMe provides a standardized evaluation for real-time procedural guidance, highlighting the gap in error detection and corrective feedback, which is crucial for practical assistants."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
