{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2607.21072", "kind": "paper", "url": "https://arxiv.org/abs/2607.21072"}, {"id": "gh:zju-omniai/provise", "kind": "repo", "url": "https://github.com/ZJU-OmniAI/ProVisE"}, {"kind": "website", "url": "https://zju-omniai.github.io/ProVisE/"}], "authors": ["Xu Wang", "Kaixiang Yao", "Miao Pan", "Xiaohe Zhou", "Xuanyu Liu", "Wenqi Zhang", "Xuhong Zhang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2607.21072", "source_url": "https://arxiv.org/abs/2607.21072"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "ProVisE is a framework for evaluating image-generation models on spatial benchmarks by converting visual answers into structured predictions. SpatialGen-Bench is a diagnostic dataset of 470 samples across 14 spatial subtasks."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2607.21072", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2607.21072", "title": "SpatialGen-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2607.21072", "modality": null, "name": "SpatialGen-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_spatialgen-bench_3d213770", "origin_source": "arxiv", "recognition_confidence": "1.0", "record_sha256": "10ba674af173046bc060eb8d954d61288ea5bc0814dbb5c54588a9e8c9cea710", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2607.21072"}, "publisher": null, "released": "2026-07-23", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2607.21072", "source_url": "https://arxiv.org/abs/2607.21072"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2607-21072", "source": "claire_radar", "source_benchmark_id": "2607.21072", "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/ZJU-OmniAI/ProVisE", "githubScope": "benchmark_repo", "githubStars": 25, "hfDailySubmittedAt": "2026-07-24T00:00:00.000Z", "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 35, "hfPaperUrl": "https://huggingface.co/papers/2607.21072", "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": {"observedAt": "2026-10-04T15:31:48.030916Z", "state": "fresh"}}, "status": "fresh"}, "capabilities": [], "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": "a7fd92873345ae54aa6f0b132092a7b892c6298b91082aa2bc60b6a21e25fd1f", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "ProVisE is a framework for evaluating image-generation models on spatial benchmarks by converting visual answers into structured predictions. SpatialGen-Bench is a diagnostic dataset of 470 samples across 14 spatial subtasks.", "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": "We further introduce SpatialGen-Bench, a curated diagnostic benchmark of 470 samples across 14 spatial subtasks, four capability levels, and diverse answer forms."}, "familyId": "bmf_54461512c988", "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/ZJU-OmniAI/ProVisE", "url": "https://github.com/ZJU-OmniAI/ProVisE"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/ZJU-OmniAI/ProVisE", "url": "https://huggingface.co/papers/2607.21072"}, {"basis": "linked-from-project", "linkedUrl": "https://github.com/ZJU-OmniAI/ProVisE", "url": "https://zju-omniai.github.io/ProVisE/"}], "scope": "project_repo", "url": "https://github.com/ZJU-OmniAI/ProVisE", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_spatialgen-bench_3d213770", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/ZJU-OmniAI/ProVisE", "data": null, "hfPaper": "https://huggingface.co/papers/2607.21072", "paper": "https://arxiv.org/abs/2607.21072", "pdf": "https://arxiv.org/pdf/2607.21072", "project": "https://zju-omniai.github.io/ProVisE/", "report": "https://arxiv.org/abs/2607.21072"}, "metrics": [], "motivation": "Spatial intelligence is essential for agents to move from static semantic understanding toward interacting with the physical world.", "name": "SpatialGen-Bench", "oneLine": "ProVisE is a framework for evaluating image-generation models on spatial benchmarks by converting visual answers into structured predictions. SpatialGen-Bench is a diagnostic dataset of 470 samples across 14 spatial subtasks.", "paperTitle": "Show, Don't Tell: Evaluating Spatial Cognition in Generative Pixels Rather Than LLM Text", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2607.21072", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "ranking": {"90d": {"confidence": "Medium", "coverage": 0.7, "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": 0.43489731988793906, "value": 25}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": 0.6331884794120033, "value": 35}}, "confidence": "Medium", "coverage": 0.7, "rank": 77, "score": 48}, "method": "attention-ranking-v13", "rank": 77, "score": 48}}, "readiness": "Runnable", "recognitionConfidence": 1.0, "relation": "introduces", "releasedAt": "2026-07-23", "source": {"authors": ["Xu Wang", "Kaixiang Yao", "Miao Pan", "Xiaohe Zhou", "Xuanyu Liu", "Wenqi Zhang", "Xuhong Zhang"], "categories": ["cs.CV"], "id": "2607.21072", "title": "Show, Don't Tell: Evaluating Spatial Cognition in Generative Pixels Rather Than LLM Text", "type": "arxiv", "url": "https://arxiv.org/abs/2607.21072"}, "sourceUpdatedAt": "2026-07-23T09:04:48Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "Existing spatial benchmarks rely on text or coordinates, limiting image-generation models. ProVisE adapts visual answers to original metrics, enabling comparison. However, unclear if it is a standalone benchmark or a framework."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
