{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.07861", "kind": "paper", "url": "https://arxiv.org/abs/2606.07861"}], "authors": ["Lujun Li", "Lama Sleem", "Niccolo Gentile", "Yangjie Xu", "Yewei Song", "Wenbo Wu", "Radu State"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.07861", "source_url": "https://arxiv.org/abs/2606.07861"}, "categories": ["General AI", "Multimodal", "Reasoning"], "description": {"en": "FineSightBench probes fine-scale visual perception in VLMs across scales from 4 to 48 pixels, separating perception and reasoning tasks."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.07861", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.07861", "title": "FineSightBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of Luxembourg", "Université Paris-Saclay"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2606.07861", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2606.07861", "source_url": "https://openalex.org/W7164017583"}, "key": "claire-radar:2606.07861", "modality": null, "name": "FineSightBench", "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_finesightbench_86226917", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "db83fdc16acbe57864c737b8f721c8916655fe2a3d30815a8b0bdb8bdbd86a77", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.07861"}, "publisher": null, "released": "2026-06-05", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.07861", "source_url": "https://arxiv.org/abs/2606.07861"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-07861", "source": "claire_radar", "source_benchmark_id": "2606.07861", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": [], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Multimodal", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": null, "githubScope": null, "githubStars": null, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2606.07861", "observedAt": "2026-10-04T03:02:42.754940Z", "signalStatus": {"githubStars": {"state": "not_applicable"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "state": "unavailable"}}, "status": "unavailable"}, "capabilities": ["Reasoning"], "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": "98af499c020f88ea65d265f60d76a812d1785747cfe993512c319c416565cc59", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "FineSightBench probes fine-scale visual perception in VLMs across scales from 4 to 48 pixels, separating perception and reasoning tasks.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence"], "snippet": "As such, we introduce FineSightBench, a new benchmark that systematically probes this limit by separating perception tasks (pixel-level recognition of letters, shapes, objects) from reasoning tasks (spatial reasoning, counting, ordering over small targets) across controlled scales of 4--48px."}, "familyId": "bmf_a5a8565d449b", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 0, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [], "status": "not-found"}, "heat": null, "id": "bm_finesightbench_86226917", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": null, "data": null, "hfPaper": "https://huggingface.co/papers/2606.07861", "paper": "https://arxiv.org/abs/2606.07861", "pdf": "https://arxiv.org/pdf/2606.07861", "project": null, "report": "https://arxiv.org/abs/2606.07861"}, "metrics": [], "motivation": "Recent vision-language models (VLMs) excel at multimodal understanding and reasoning, yet their fine-grained visual perception remains underexplored.", "name": "FineSightBench", "oneLine": "FineSightBench probes fine-scale visual perception in VLMs across scales from 4 to 48 pixels, separating perception and reasoning tasks.", "paperTitle": "The Last Visible Pixel: Probing Fine-Scale Perception in Vision-Language Models", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.07861", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {}, "readiness": "Paper only", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-05", "source": {"authors": ["Lujun Li", "Lama Sleem", "Niccolo Gentile", "Yangjie Xu", "Yewei Song", "Wenbo Wu and Radu State"], "categories": ["cs.CV", "cs.AI"], "id": "2606.07861", "title": "The Last Visible Pixel: Probing Fine-Scale Perception in Vision-Language Models", "type": "arxiv", "url": "https://arxiv.org/abs/2606.07861"}, "sourceUpdatedAt": "2026-06-05T21:49:34Z", "topics": ["Multimodal", "Reasoning"], "venueAttempts": [], "whyItMatters": "Reveals limits in fine-grained VLM perception, motivating better evaluation, but is a probe with no public path."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
