{"record": {"aliases": ["Human-Centric Image Captioning with Subject-Centered Spatial Understanding"], "artifacts": [{"id": "arxiv:2609.08300", "kind": "paper", "url": "https://arxiv.org/abs/2609.08300"}, {"id": "gh:jhang2020/space-eval", "kind": "repo", "url": "https://github.com/JHang2020/SPACE-Eval"}], "authors": ["Bozhou Li", "Jiahang Zhang", "Yue Ding", "Yushuo Guan", "Bohan Zeng", "Yiyan Ji", "Xinlong Chen", "Yang Shi", "Yifan Dai", "Yuran Wang", "Chengzhuo Tong", "Pengfei Wan", "Yuanxing Zhang", "Wentao Zhang"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2609.08300", "source_url": "https://arxiv.org/abs/2609.08300"}, "categories": ["General AI", "Vision & 3D", "cs.CV"], "description": {"en": "SPACE evaluates subject-centered spatial understanding in human-centric image captioning, focusing on egocentric laterality and anatomical-object bindings."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2609.08300", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2609.08300", "title": "SPACE"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "institutions": ["Peking University", "Chinese Academy of Sciences", "Nanjing University", "Shanghai Jiao Tong University"], "institutions_reference": {"basis": "openalex_linked_institutions", "doi": "10.48550/arxiv.2609.08300", "retrieved_at": "2026-11-04", "source_key": "claire-radar:2609.08300", "source_url": "https://openalex.org/W7212083729"}, "key": "claire-radar:2609.08300", "modality": null, "name": "SPACE", "openness": {"code_license": null, "data_license": null, "evidence": [], "status": "unknown"}, "provenance": {"admission_policy_version": "2026-09-09.1", "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_human-centric-image-captioning-with-subjec_25bb01f1", "origin_source": "arxiv", "recognition_confidence": "0.7", "record_sha256": "7d8eab3c60f78e46ecfe6508650d0af1215035eaf4e57c9a8ea6832730c43724", "relation": "introduces", "review_model": "deepseek-v4-pro", "review_state": "ai-reviewed", "reviewed_at": "2026-09-10T00:52:28.558291Z", "source_url": "https://arxiv.org/abs/2609.08300"}, "publisher": null, "released": "2026-09-08", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2609.08300", "source_url": "https://arxiv.org/abs/2609.08300"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2609-08300", "source": "claire_radar", "source_benchmark_id": "2609.08300", "source_metadata": {"claire_radar": {"adoption30d": null, "aliases": ["Human-Centric Image Captioning with Subject-Centered Spatial Understanding"], "annotation": "Unknown", "applicationDomains": ["General AI"], "area": "Vision & 3D", "attention": {"asOf": "2026-09-09", "attemptedAt": "2026-09-09T06:48:27.347268Z", "githubRepo": "https://github.com/JHang2020/SPACE-Eval", "githubScope": "benchmark_repo", "githubStars": 1, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": null, "hfPaperUrl": "https://huggingface.co/papers/2609.08300", "observedAt": "2026-09-09T06:48:27.347268Z", "signalStatus": {"githubStars": {"observedAt": "2026-09-09T06:48:27.347268Z", "state": "fresh"}, "hfDatasetDownloads": {"state": "not_applicable"}, "hfDatasetLikes": {"state": "not_applicable"}, "hfPaperUpvotes": {"checkedAt": "2026-09-09T06:48:27.347268Z", "state": "unavailable"}}, "status": "fresh"}, "attentionForecast": {"confidence": "Medium", "horizon": "7d", "reason": "Human-centric spatial grounding is an active area for multimodal models, and the benchmark includes a code link, which improves discoverability.", "score": 40}, "capabilities": [], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-09-10T00:52:28.558291Z", "inputHash": "07f419f835b5c9efbef141e4e716caa7b22ced1f7ab9b1c5617124fe1d108043", "model": "deepseek-v4-pro"}, "curation": {"admissionPolicyVersion": "2026-09-09.1", "canonicalNameEvidence": "SPACE (Subject-centric Poses, Appearance, and Characteristics Evaluation), a benchmark designed to evaluate subject-centered spatial understanding.", "canonicalNameSource": "abstract", "decisionReason": "SPACE is a formally named benchmark with code availability, a defined evaluation focus, and experimental usage.", "evaluationEvidence": {"reuse": "Our benchmark and code are available at https://github.com/JHang2020/SPACE-Eval.", "scoring": "Extensive experiments on SPACE demonstrate our framework significantly improves human-centric caption quality, particularly in subject-centered spatial reasoning, achieving performance competitive with strong closed-source models.", "task": "SPACE (Subject-centric Poses, Appearance, and Characteristics Evaluation), a benchmark designed to evaluate subject-centered spatial understanding."}, "model": "deepseek-v4-pro", "reviewedAt": "2026-09-10T00:52:28.558291Z", "state": "ai-reviewed"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SPACE evaluates subject-centered spatial understanding in human-centric image captioning, focusing on egocentric laterality and anatomical-object bindings.", "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": "To systematically expose and quantify this bottleneck, we introduce SPACE (Subject-centric Poses, Appearance, and Characteristics Evaluation), a benchmark designed to evaluate subject-centered spatial understanding."}, "familyId": "bmf_930467cec06f", "firstSeenAt": "2026-09-09", "heat": null, "id": "bm_human-centric-image-captioning-with-subjec_25bb01f1", "indexedAt": "2026-09-09T06:18:29.794093Z", "industrySectors": [], "links": {"code": "https://github.com/JHang2020/SPACE-Eval", "data": null, "paper": "https://arxiv.org/abs/2609.08300", "pdf": "https://arxiv.org/pdf/2609.08300", "project": null, "report": "https://arxiv.org/abs/2609.08300"}, "metrics": [], "motivation": "While multimodal large language models (MLLMs) achieve remarkable performance on generic image captioning, they frequently suffer from structural hallucinations in human-centric scenarios.", "name": "SPACE", "oneLine": "SPACE evaluates subject-centered spatial understanding in human-centric image captioning, focusing on egocentric laterality and anatomical-object bindings.", "paperTitle": "Human-Centric Image Captioning with Subject-Centered Spatial Understanding", "primaryDomain": "General AI", "publication": {"evidence": null, "evidenceLevel": "unverified", "evidenceUrl": "https://arxiv.org/abs/2609.08300", "source": "arxiv-metadata", "status": "unverified", "venue": null, "verifiedAt": "2026-09-09T06:18:29.794093Z"}, "publications": [], "ranking": {"30d": {"confidence": "Low", "coverage": 0.55, "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": 30}, "level": {"components": {"githubStars": {"normalization": "log-max", "normalized": 0.10323645750017778, "value": 1}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 99, "score": 28}, "method": "attention-ranking-v13", "rank": 99, "score": 28}, "90d": {"confidence": "Low", "coverage": 0.55, "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": 0.09361440452337855, "value": 1}, "hfDatasetDownloads": {"normalization": "log-max", "normalized": null, "value": null}, "hfPaperUpvotes": {"normalization": "log-max", "normalized": null, "value": null}}, "confidence": "Low", "coverage": 0.55, "rank": 256, "score": 28}, "method": "attention-ranking-v13", "rank": 256, "score": 28}, "today": {"confidence": "Low", "coverage": 0.25, "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": 1, "historyDate": "2026-09-08", "rank": null, "score": null, "windowDays": 1}, "level": {"components": {"githubStars": {"normalization": "percentile", "normalized": 0.6428571428571429, "percentile": 0.6428571428571429, "value": 1}, "hfDatasetDownloads": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "hfPaperUpvotes": {"normalization": "percentile", "normalized": null, "percentile": null, "value": null}, "llmAttentionForecast": {"maxContribution": 0.25, "percentile": 0.6, "role": "bonus", "value": 60}}, "confidence": "Low", "coverage": 0.25, "rank": 2, "score": 56}, "method": "attention-ranking-v13", "rank": 2, "score": 56}}, "readiness": "Runnable", "recognitionConfidence": 0.7, "relation": "introduces", "releasedAt": "2026-09-08", "source": {"authors": ["Bozhou Li", "Jiahang Zhang", "Yue Ding", "Yushuo Guan", "Bohan Zeng", "Yiyan Ji", "Xinlong Chen", "Yang Shi", "Yifan Dai", "Yuran Wang", "Chengzhuo Tong", "Pengfei Wan", "Yuanxing Zhang", "Wentao Zhang"], "categories": ["cs.CV"], "id": "2609.08300", "title": "Human-Centric Image Captioning with Subject-Centered Spatial Understanding", "type": "arxiv", "url": "https://arxiv.org/abs/2609.08300"}, "sourceUpdatedAt": "2026-09-08T06:21:32Z", "topics": ["cs.CV"], "venueAttempts": [], "whyItMatters": "It exposes structural hallucinations in human-centric captions that undermine downstream applications like avatar generation and action understanding."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
