{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.08894", "kind": "paper", "url": "https://arxiv.org/abs/2606.08894"}, {"id": "gh:yizheng-sun/distract-bench", "kind": "repo", "url": "https://github.com/Yizheng-Sun/Distract-Bench"}], "authors": ["Yizheng Sun", "Mochuan Zhan", "Yanan Ma", "Jia Tong See", "Yifan Wang", "Ziyi Wang", "Hao Li", "Yang Cui", "Wenhao Cai", "Jingyu Sun", "Chenghua Lin", "Riza Batista-Navarro", "Jingyuan Sun"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.08894", "source_url": "https://arxiv.org/abs/2606.08894"}, "categories": ["General AI", "Multimodal", "Reasoning", "Robustness", "Safety & Trustworthiness"], "description": {"en": "Distract-Bench evaluates robustness of vision-language models to semantic visual distractions, which are meaningful but task-irrelevant cues that preserve the ground-truth answer."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.08894", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.08894", "title": "Distract-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of Manchester", "Marex (Norway)", "Imperial College London"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.08894", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.08894", "source_url": "https://openalex.org/W7164003128"}, "key": "claire-radar:2606.08894", "modality": null, "name": "Distract-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_distract-bench_3873c48a", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "03785629d596a7ce1217d1a140c14d3880f6ea6f746b3dd102bbfb59563c76bd", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.08894"}, "publisher": null, "released": "2026-06-08", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.08894", "source_url": "https://arxiv.org/abs/2606.08894"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-08894", "source": "claire_radar", "source_benchmark_id": "2606.08894", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Safety & Trustworthiness", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/Yizheng-Sun/Distract-Bench", "githubScope": "benchmark_repo", "githubStars": 10, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2606.08894", "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"}, "capabilities": ["Reasoning", "Robustness"], "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": "3a746734efbdf9199a373fc70cc1f55de7066e6c16c76f7437758e552168964d", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Distract-Bench evaluates robustness of vision-language models to semantic visual distractions, which are meaningful but task-irrelevant cues that preserve the ground-truth answer.", "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 address this gap, we introduce \\textbf{Distract-Bench}, a benchmark for evaluating VLM robustness to \\textbf{semantic visual distractions}, defined as meaningful but task-irrelevant visual cues added to inputs while preserving the ground-truth answer."}, "familyId": "bmf_74ca191db947", "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/Yizheng-Sun/Distract-Bench", "url": "https://github.com/Yizheng-Sun/Distract-Bench"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/Yizheng-Sun/Distract-Bench", "url": "https://huggingface.co/papers/2606.08894"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/Yizheng-Sun/Distract-Bench", "url": "https://arxiv.org/abs/2606.08894"}, {"basis": "linked-from-report", "linkedUrl": "https://github.com/Yizheng-Sun/Distract-Bench.", "url": "https://arxiv.org/abs/2606.08894"}], "scope": "benchmark_repo", "url": "https://github.com/Yizheng-Sun/Distract-Bench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_distract-bench_3873c48a", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/Yizheng-Sun/Distract-Bench", "data": null, "hfPaper": "https://huggingface.co/papers/2606.08894", "paper": "https://arxiv.org/abs/2606.08894", "pdf": "https://arxiv.org/pdf/2606.08894", "project": null, "report": "https://arxiv.org/abs/2606.08894"}, "metrics": [], "motivation": "Reasoning Vision-Language Models (VLMs) achieve strong performance on complex multimodal tasks, but reliable real-world application requires handling visual inputs that are messier than clean, curated benchmarks.", "name": "Distract-Bench", "oneLine": "Distract-Bench evaluates robustness of vision-language models to semantic visual distractions, which are meaningful but task-irrelevant cues that preserve the ground-truth answer.", "paperTitle": "Are Reasoning Vision-Language Models Robust to Semantic Visual Distractions?", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.08894", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "Yizheng-Sun", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://github.com/Yizheng-Sun/Distract-Bench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-08", "source": {"authors": ["Yizheng Sun", "Mochuan Zhan", "Yanan Ma", "Jia Tong See", "Yifan Wang", "Ziyi Wang", "Hao Li", "Yang Cui", "Wenhao Cai", "Jingyu Sun", "Chenghua Lin", "Riza Batista-Navarro", "Jingyuan Sun"], "categories": ["cs.CV", "cs.CL"], "id": "2606.08894", "title": "Are Reasoning Vision-Language Models Robust to Semantic Visual Distractions?", "type": "arxiv", "url": "https://arxiv.org/abs/2606.08894"}, "sourceUpdatedAt": "2026-06-08T00:40:32Z", "topics": ["Multimodal", "Reasoning"], "venueAttempts": [], "whyItMatters": "It exposes a distinct failure mode where models perceive evidence correctly but reason from distracting cues, shifting robustness evaluation from perceptual degradation to distraction handling."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
