{"record": {"aliases": ["NumerosityVLM: A Cognitively Inspired Benchmark for Interpreting Numerosity Representations in Vision-Language Models"], "artifacts": [{"id": "arxiv:2608.15425", "kind": "paper", "url": "https://arxiv.org/abs/2608.15425"}, {"id": "gh:fuy3/numerosityvlm-benchmark", "kind": "repo", "url": "https://github.com/fuy3/NumerosityVLM-Benchmark"}, {"id": "hf:fuy3/numerosityvlm", "kind": "dataset", "url": "https://huggingface.co/datasets/fuy3/NumerosityVLM"}], "authors": ["Yiming Fu", "Fangjun Li", "Xiujin Liu", "Ruidong Ma", "Hang Yu", "Zhichen Lu", "Kanwei He", "Alessandro Di Nuovo", "Angelo Cangelosi", "Zhegong Shangguan"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2608.15425", "source_url": "https://arxiv.org/abs/2608.15425"}, "categories": ["General AI", "Multimodal"], "description": {"en": "Evaluates zero-shot numerosity perception in vision-language models on 10,800 synthetic images across six controlled conditions manipulating object size, spatial arrangement, and numerosity while ablating texture, shape, and color."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2608.15425", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2608.15425", "title": "NumerosityVLM"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["University of Manchester", "University of Michigan", "Sheffield Hallam University", "Tufts University", "École Nationale Supérieure de Techniques Avancées", "Institut Polytechnique de Paris", "École Nationale Supérieure de Techniques Avancées Paris"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2608.15425", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2608.15425", "source_url": "https://openalex.org/W7203709002"}, "key": "claire-radar:2608.15425", "modality": null, "name": "NumerosityVLM", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"confidence": "Medium", "crawl_bundle": "claire_radar_2026-10-14", "crawled_at": "2026-10-14T09:44:51Z", "data_status": "primary-source-indexed", "display_eligible": "true", "origin_record_id": "bm_numerosityvlm-a-cognitively-inspired-bench_6f2495fb", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "f655b80f1e480cfa23c055903c510cd1d16fd7744f224306006c5dc1ebdaa785", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-08-25T23:23:46.758372Z", "source_url": "https://arxiv.org/abs/2608.15425"}, "publisher": null, "released": "2026-08-15", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2608.15425", "source_url": "https://arxiv.org/abs/2608.15425"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2608-15425", "source": "claire_radar", "source_benchmark_id": "2608.15425", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["NumerosityVLM: A Cognitively Inspired Benchmark for Interpreting Numerosity Representations in Vision-Language Models"], "annotation": "Machine Generated", "applicationDomains": ["General AI"], "area": "Multimodal", "attention": {"asOf": "2026-10-05", "attemptedAt": "2026-10-04T15:31:48.030916Z", "githubRepo": "https://github.com/fuy3/NumerosityVLM-Benchmark", "githubScope": "benchmark_repo", "githubStars": 1, "hfDailySubmittedAt": null, "hfDataset": "fuy3/NumerosityVLM", "hfDatasetDownloads": 54, "hfDatasetLikes": 2, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2608.15425", "observedAt": "2026-10-04T15:31:48.030916Z", "signalStatus": {"githubStars": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}, "hfDatasetDownloads": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}, "hfDatasetLikes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-09-30", "state": "stale"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "The benchmark addresses a niche but active area of vision-language model interpretability with public code and data, likely attracting moderate attention from cognitive AI researchers.", "score": 35}, "capabilities": [], "confidence": "Medium", "construction": "Original Synthetic", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-25T23:23:46.758372Z", "inputHash": "c7ae0054ca6e392a9d939120348a5436a91f35e41953ae9ccfa7556fa0bb7b82", "model": "deepseek-v4-pro"}, "curation": {"canonicalNameEvidence": "We introduce a cognitively inspired diagnostic benchmark, NumerosityVLM, comprising 10,800 synthetic images across six controlled conditions.", "canonicalNameSource": "abstract", "decisionReason": "The benchmark is formally named, provides code and data artifacts for reuse, and defines a fixed evaluation protocol with zero-shot scoring.", "model": "deepseek-v4-pro", "reviewedAt": "2026-08-25T23:23:46.758372Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "Evaluates zero-shot numerosity perception in vision-language models on 10,800 synthetic images across six controlled conditions manipulating object size, spatial arrangement, and numerosity while ablating texture, shape, and color.", "displayEligible": true, "domainCuration": {"method": "rules-v1", "state": "auto"}, "evaluationMode": "public_reusable", "evidence": {"reasonCodes": ["benchmark term in abstract", "benchmark release stated in one sentence", "evaluation protocol evidence", "public artifact URL"], "snippet": "We introduce a cognitively inspired diagnostic benchmark, NumerosityVLM, comprising 10,800 synthetic images across six controlled conditions."}, "familyId": "bmf_804131cc2b86", "firstSeenAt": "2026-08-19", "heat": null, "id": "bm_numerosityvlm-a-cognitively-inspired-bench_6f2495fb", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/fuy3/NumerosityVLM-Benchmark", "data": "https://huggingface.co/datasets/fuy3/NumerosityVLM", "paper": "https://arxiv.org/abs/2608.15425", "pdf": "https://arxiv.org/pdf/2608.15425", "project": null, "report": "https://arxiv.org/abs/2608.15425"}, "metrics": [], "motivation": "Vision-language models (VLMs) achieve strong performance on high-level multimodal tasks, yet numerosity perception, a cognitive ability that emerges in human infants before language acquisition, remains poorly understood in current models, as existing counting benchmarks entangle numerosity with correlated visual factors.", "name": "NumerosityVLM", "oneLine": "Evaluates zero-shot numerosity perception in vision-language models on 10,800 synthetic images across six controlled conditions manipulating object size, spatial arrangement, and numerosity while ablating texture, shape, and color.", "paperTitle": "NumerosityVLM: A Cognitively Inspired Benchmark for Interpreting Numerosity Representations in Vision-Language Models", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2608.15425", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-08-19T11:05:13.395391Z"}, "publications": [], "publishers": [{"name": "fuy3", "organizationType": "community", "role": "benchmark-publisher", "sourceUrl": "https://github.com/fuy3/NumerosityVLM-Benchmark"}], "ranking": {"90d": {"confidence": "High", "coverage": 1.0, "datasetDownloadRank": 77, "datasetRankPopulation": 85, "growth": {"components": {"githubStars": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}}, "confidence": "Low", "coverage": 0.0, "elapsedDays": null, "historyDate": null, "rank": null, "score": null, "windowDays": 90}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.09252268850709358, "value": 1}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": 0.4126748623835525, "value": 54}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.0, "value": 0}}, "confidence": "High", "coverage": 1.0, "rank": 565, "score": 17}, "method": "attention-ranking-v13", "rank": 565, "score": 17}}, "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-08-15", "source": {"authors": ["Yiming Fu", "Fangjun Li", "Xiujin Liu", "Ruidong Ma", "Hang Yu", "Zhichen Lu", "Kanwei He", "Alessandro Di Nuovo", "Angelo Cangelosi", "Zhegong Shangguan"], "categories": ["cs.CV", "cs.AI"], "id": "2608.15425", "title": "NumerosityVLM: A Cognitively Inspired Benchmark for Interpreting Numerosity Representations in Vision-Language Models", "type": "arxiv", "url": "https://arxiv.org/abs/2608.15425"}, "sourceUpdatedAt": "2026-08-15T22:01:21Z", "topics": ["Multimodal"], "venueAttempts": [], "whyItMatters": "Isolates numerosity from correlated visual features, enabling targeted diagnosis of number understanding in VLMs and revealing that architecture-driven language components limit performance."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
