{"paper":{"title":"SceneForge: Structured World Supervision from 3D Interventions","license":"http://creativecommons.org/licenses/by/4.0/","headline":"SceneForge generates consistent supervision for removal tasks by propagating explicit interventions through editable 3D scene states.","cross_cats":["cs.GR"],"primary_cat":"cs.CV","authors_text":"Danny Wicks, Jiayang Ao, Jizhizi Li, Petru-Daniel Tudosiu","submitted_at":"2026-05-14T05:38:00Z","abstract_excerpt":"Many multimodal learning tasks require supervision that remains consistent across edits, viewpoints, and scene-level interventions. However, such supervision is difficult to obtain from observation-level datasets, which do not expose the underlying scene state or how changes propagate through it. We present SceneForge, an intervention-driven framework that generates structured supervision from editable 3D world states. SceneForge represents each scene as a persistent world with semantic, geometric, and physical dependencies. By applying explicit interventions (e.g., object removal or camera va"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Under matched training budgets, incorporating SceneForge supervision improves both object removal and scene removal performance across multiple benchmarks in both quantitative and qualitative evaluation.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That supervision generated from synthetic 3D interventions transfers effectively to real-world images and that the modeled scene dependencies accurately reflect real physical and geometric relationships.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SceneForge creates intervention-consistent multimodal supervision from editable 3D world states, yielding improved object and scene removal performance on benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SceneForge generates consistent supervision for removal tasks by propagating explicit interventions through editable 3D scene states.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"817753febd56f6518c2b5c7ec25a9068d8534f042e3c441fcc0e50c2645c49b3"},"source":{"id":"2605.14399","kind":"arxiv","version":1},"verdict":{"id":"270faa76-6a41-45d1-a709-62f304ade94f","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:30:14.318177Z","strongest_claim":"Under matched training budgets, incorporating SceneForge supervision improves both object removal and scene removal performance across multiple benchmarks in both quantitative and qualitative evaluation.","one_line_summary":"SceneForge creates intervention-consistent multimodal supervision from editable 3D world states, yielding improved object and scene removal performance on benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That supervision generated from synthetic 3D interventions transfers effectively to real-world images and that the modeled scene dependencies accurately reflect real physical and geometric relationships.","pith_extraction_headline":"SceneForge generates consistent supervision for removal tasks by propagating explicit interventions through editable 3D scene states."},"references":{"count":38,"sample":[{"doi":"","year":null,"title":"Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , year =","work_id":"4a1b321e-0ca7-4aec-92bc-e5aaf03c1f2c","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Proceedings of the IEEE/CVF International Conference on Computer Vision , year=","work_id":"307a52fd-3187-4fca-aa4a-27a544329543","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , year =","work_id":"73025e16-adf2-410a-b718-1c4360a49bf0","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=","work_id":"ad8d5f1e-c429-41f8-bcc3-aeeb57ab2c0c","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":null,"title":"Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=","work_id":"54619911-71f5-4f08-b8be-216bb9a80fe5","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":38,"snapshot_sha256":"02438074d8628649fb28ff70251accb8faef1b2d61155c716342ccd3b309aeb4","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}