{"record": {"aliases": [], "artifacts": [{"id": "arxiv:2606.30124", "kind": "paper", "url": "https://arxiv.org/abs/2606.30124"}, {"id": "gh:mair-lab-hust/sciir", "kind": "repo", "url": "https://github.com/MAIR-Lab-HUST/SciIR"}], "authors": ["Zhiyuan Ma", "Zhengfeng Shi", "Yuning An", "Peize Li", "Jiabao Wei", "Ruijie Li", "Junhao Xiao", "Jianjun Li", "Bowen Zhou"], "authors_reference": {"basis": "arxiv_metadata", "retrieved_at": "2026-11-03", "source_key": "claire-radar:2606.30124", "source_url": "https://arxiv.org/abs/2606.30124"}, "categories": ["General AI", "Reasoning", "Vision & 3D"], "description": {"en": "SciIR-Bench evaluates text-to-image models on scientific image reasoning across three semiotic-aligned tracks: entity structure, scientific process, and scientific law. It uses an atomic checklist to convert scientific accuracy into verifiable fine-grained questions."}, "documents": [{"document_type": "registry_page", "id": "claire_radar:https://arxiv.org/abs/2606.30124", "source": "claire_radar", "source_url": "https://arxiv.org/abs/2606.30124", "title": "SciIR-Bench"}], "evidence_summary": {"document_count": 1, "model_count": null, "model_count_basis": null}, "key": "claire-radar:2606.30124", "modality": null, "name": "SciIR-Bench", "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": "unknown", "origin_record_id": "bm_sciir-bench_0a7d1ae8", "origin_source": "arxiv", "recognition_confidence": "0.85", "record_sha256": "7290b0f5b18c406280aaaf57d93f7ba59d32eab1f89edd2b074193404a6c35d8", "relation": "introduces", "review_state": "unreviewed", "source_url": "https://arxiv.org/abs/2606.30124"}, "publisher": null, "released": "2026-06-29", "released_reference": {"basis": "benchmark_release", "source_key": "claire-radar:2606.30124", "source_url": "https://arxiv.org/abs/2606.30124"}, "schema_version": 1, "sizes": [], "slug": "claire-radar-2606-30124", "source": "claire_radar", "source_benchmark_id": "2606.30124", "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/MAIR-Lab-HUST/SciIR", "githubScope": "benchmark_repo", "githubStars": 7, "hfDailySubmittedAt": null, "hfDataset": null, "hfDatasetDownloads": null, "hfDatasetLikes": null, "hfPaperUpvotes": 5, "hfPaperUrl": "https://huggingface.co/papers/2606.30124", "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": {"checkedAt": "2026-10-04T15:31:48.030916Z", "lastSuccessfulDate": "2026-10-04", "state": "stale"}}, "status": "fresh"}, "capabilities": ["Reasoning"], "confidence": "Medium", "construction": "Unknown", "constructionDetail": "Unknown — the indexer does not infer construction details without explicit source evidence.", "copyGeneration": {"generatedAt": "2026-08-20T15:22:10.044731Z", "inputHash": "83d731dbb3f312a2f5e2092f2b709ab7f0034d8f9dd4ec634cb4c2f8520ef193", "model": "deepseek-v4-flash"}, "dataStatus": "primary-source-indexed", "demo": false, "description": "SciIR-Bench evaluates text-to-image models on scientific image reasoning across three semiotic-aligned tracks: entity structure, scientific process, and scientific law. It uses an atomic checklist to convert scientific accuracy into verifiable fine-grained questions.", "domainCuration": {"method": "rules-v1", "state": "auto"}, "evidence": {"reasonCodes": ["exact named benchmark artifact released in abstract", "benchmark release stated in one sentence"], "snippet": "For evaluation, we propose SciIR-Bench, which aligns with these three semiotic levels and employs an Atomic Checklist to convert the outcome-oriented scientific accuracy into process-oriented, verifiable, fine-grained questions."}, "familyId": "bmf_e6067af2051a", "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/MAIR-Lab-HUST/SciIR", "url": "https://github.com/MAIR-Lab-HUST/SciIR"}, {"basis": "existing-metric-association", "linkedUrl": "https://github.com/MAIR-Lab-HUST/SciIR", "url": "https://huggingface.co/papers/2606.30124"}], "scope": "benchmark_repo", "url": "https://github.com/MAIR-Lab-HUST/SciIR", "verification": "source-linked"}], "status": "linked"}, "heat": null, "id": "bm_sciir-bench_0a7d1ae8", "indexedAt": "2026-08-19T11:05:13.395391Z", "industrySectors": [], "links": {"code": "https://github.com/MAIR-Lab-HUST/SciIR", "data": null, "hfPaper": "https://huggingface.co/papers/2606.30124", "paper": "https://arxiv.org/abs/2606.30124", "pdf": "https://arxiv.org/pdf/2606.30124", "project": null, "report": "https://arxiv.org/abs/2606.30124"}, "metrics": [], "motivation": "While Text-to-Image (T2I) models have shown remarkable success in generating photorealistic visual content, they still struggle with the rigorous semantic alignment and logical reasoning required for scientific imagery.", "name": "SciIR-Bench", "oneLine": "SciIR-Bench evaluates text-to-image models on scientific image reasoning across three semiotic-aligned tracks: entity structure, scientific process, and scientific law. It uses an atomic checklist to convert scientific accuracy into verifiable fine-grained questions.", "paperTitle": "SciIR: A Large-scale Training Dataset and Benchmark for Scientific Image Reasoning Generation", "primaryDomain": "General AI", "publication": {"evidence": "Accepted to ECCV 2026", "evidenceLevel": "author-claim", "evidenceUrl": "https://arxiv.org/abs/2606.30124", "source": "arxiv-comments", "status": "acceptance_claimed", "venue": "ECCV 2026", "verifiedAt": "2026-10-14T01:46:58.058199Z"}, "publications": [], "publishers": [{"name": "MAIR-Lab-HUST", "organizationType": "academic-lab", "role": "benchmark-publisher", "sourceUrl": "https://github.com/MAIR-Lab-HUST/SciIR"}], "ranking": {}, "readiness": "Runnable", "recognitionConfidence": 0.85, "relation": "introduces", "releasedAt": "2026-06-29", "source": {"authors": ["Zhiyuan Ma", "Zhengfeng Shi", "Yuning An", "Peize Li", "Jiabao Wei", "Ruijie Li", "Junhao Xiao", "Jianjun Li", "Bowen Zhou"], "categories": ["cs.CV"], "id": "2606.30124", "title": "SciIR: A Large-scale Training Dataset and Benchmark for Scientific Image Reasoning Generation", "type": "arxiv", "url": "https://arxiv.org/abs/2606.30124"}, "sourceUpdatedAt": "2026-06-29T10:59:10Z", "topics": ["Reasoning"], "venueAttempts": [{"decisionRaw": "Accepted to ECCV 2026", "evidence": [{"level": "author-claim", "observedAt": "2026-10-14T01:46:58.058199Z", "rawValue": "Accepted to ECCV 2026", "sourceType": "arxiv-comments", "sourceUrl": "https://arxiv.org/abs/2606.30124"}], "reviewStatus": "accepted", "venueName": "ECCV 2026"}], "whyItMatters": "Current text-to-image models lack rigorous evaluation for scientific imagery, which requires logical reasoning beyond visual fidelity. SciIR-Bench provides a structured protocol to measure such capabilities, aiding selection and development of models for scientific visualization tasks."}}}, "schema_version": 1, "scores_by_source": {}, "siblings": []}
