{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.22476", "kind": "paper", "url": "https://arxiv.org/abs/2606.22476"}, {"id": "gh:earth-insights/cvsbench", "kind": "repo", "url": "https://github.com/earth-insights/CVSBench"}, {"id": "hf:zlyzlyzly/cvsbench", "kind": "dataset", "url": "https://huggingface.co/datasets/zlyzlyzly/CVSBench"}], "authors": ["Ruixun Liu", "Lingyu Zhang", "Lanxuan Xue", "Kaiyu Li", "Bowen Fu", "Xiangyong Cao"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.22476", "source_url": "https://arxiv.org/abs/2606.22476"}, "categories": ["General AI", "Reasoning", "Vision & 3D"], "description": {"en": "CVSBench evaluates cross-view spatial reasoning in vision-language models using satellite-street image pairs. It includes tasks for cross-view VQA, grounding, and viewpoint identification, with 3,297 image groups, 9,468 object-level annotations, and 40,679 QA pairs."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.22476", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.22476", "title": "CVSBench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.22476", "modality": null, "name": "CVSBench", "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_cvsbench_ac22ff1a", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "b8dbf0a13b0d119859056587e8533a599cbacab92a7e25088651956aa6f295db", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.22476"}, "publisher": null, "released": "2026-06-21", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.22476", "source_url": "https://arxiv.org/abs/2606.22476"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-22476", "source": "claire_radar", "source_benchmark_id": "2606.22476", "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/earth-insights/CVSBench", "githubScope": "benchmark_repo", "githubStars": 12, "hfDailySubmittedAt": null, "hfDataset": "zlyzlyzly/CVSBench", "hfDatasetDownloads": 774, "hfDatasetLikes": 1, "hfPaperUpvotes": 0, "hfPaperUrl": "https://huggingface.co/papers/2606.22476", "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": "2026-10-04", "state": "stale"}, "hfDatasetLikes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-04", "state": "stale"}, "hfPaperUpvotes": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-04", "state": "stale"}}, "status": "fresh"}, "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": "62431a688503ade3ff59c61243b231fb5e9b0ff0aa9e6c6c754b8b2e5d0a2e4a", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "CVSBench evaluates cross-view spatial reasoning in vision-language models using satellite-street image pairs. It includes tasks for cross-view VQA, grounding, and viewpoint identification, with 3,297 image groups, 9,468 object-level annotations, and 40,679 QA pairs.", "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": "Motivated by this, we introduce CVSBench, a large-scale benchmark for evaluating cross-view spatial reasoning through satellite-street pairs."}, "familyId": "bmf_5b2d8ea2ca58", "firstSeenAt": "2026-08-19", "githubIndex": {"candidateCount": 2, "indexedAt": "2026-09-17T04:40:34.111665+00:00", "repositories": [{"evidence": [{"basis": "linked-from-data", "linkedUrl": "https://github.com/earth-insights/CVSBench", "url": "https://huggingface.co/datasets/zlyzlyzly/CVSBench"}], "scope": "benchmark_repo", "url": "https://github.com/earth-insights/CVSBench", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_cvsbench_ac22ff1a", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/earth-insights/CVSBench", "data": "https://huggingface.co/datasets/zlyzlyzly/CVSBench", "hfPaper": "https://huggingface.co/papers/2606.22476", "paper": "https://arxiv.org/abs/2606.22476", "pdf": "https://arxiv.org/pdf/2606.22476", "project": null, "report": "https://arxiv.org/abs/2606.22476"}, "metrics": [], "motivation": "Humans can effortlessly reason about scenes across different viewpoints, yet it remains unclear whether Vision-Language Models (VLMs) possess similar cross-view spatial abilities.", "name": "CVSBench", "oneLine": "CVSBench evaluates cross-view spatial reasoning in vision-language models using satellite-street image pairs. It includes tasks for cross-view VQA, grounding, and viewpoint identification, with 3,297 image groups, 9,468 object-level annotations, and 40,679 QA pairs.", "paperTitle": "CVSBench: A Comprehensive Benchmark for Cross-view Spatial Reasoning and Dreaming", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2606.22476", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "earth-insights", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://huggingface.co/datasets/zlyzlyzly/CVSBench"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-06-21", "source": {"authors": ["Ruixun Liu", "Lingyu Zhang", "Lanxuan Xue", "Kaiyu Li", "Bowen Fu", "Xiangyong Cao"], "categories": ["cs.CV"], "id": "2606.22476", "title": "CVSBench: A Comprehensive Benchmark for Cross-view Spatial Reasoning and Dreaming", "type": "arxiv", "url": "https://arxiv.org/abs/2606.22476"}, "sourceUpdatedAt": "2026-06-21T12:35:43Z", "topics": ["Reasoning"], "venueAttempts": [], "whyItMatters": "CVSBench addresses the gap in evaluating VLMs' ability to reason about scenes across drastically different viewpoints, which is crucial for applications like navigation and remote sensing. It provides a systematic protocol for measuring object-level and layout consistency under viewpoint changes."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
