{"paper":{"title":"CRePE: Curved Ray Expectation Positional Encoding for Unified-Camera-Controlled Video Generation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"CRePE represents each image token as a depth-aware positional distribution along its source ray to support unified camera control under the Unified Camera Model.","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CV","authors_text":"Jong Chul Ye, Seonghyun Jin, Sunwoo Park, Youngmin Kim","submitted_at":"2026-05-13T03:18:26Z","abstract_excerpt":"Camera-conditioned video generation requires positional encoding that remains reliable under changes in camera motion, lens configuration, and scene structure. However, existing attention-level camera encodings either provide ray-only camera signals or rely on pinhole camera geometry, limiting their applicability to general camera control under the Unified Camera Model, including wide-angle and fisheye lenses. To address this limitation, we propose Curved Ray Expectation Positional Encoding (CRePE). CRePE represents each image token as a depth-aware positional distribution along its source ray"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"CRePE represents each image token as a depth-aware positional distribution along its source ray, providing a Unified Camera Model-compatible positional encoding that captures the projected-path geometry induced by wide-angle and fisheye cameras.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That pseudo-supervision from a monocular geometry foundation model is sufficient to stabilize the Geometric Attention Adapter without introducing systematic bias in the learned ray distributions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"CRePE supplies depth-aware positional distributions along curved rays for stable unified-camera control in frozen video DiT models.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"CRePE represents each image token as a depth-aware positional distribution along its source ray to support unified camera control under the Unified Camera Model.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"632fe6cff5c5ae1917b661fb5c1bc316ad5d8d6a7ce1374b0118791e84699e75"},"source":{"id":"2605.12938","kind":"arxiv","version":1},"verdict":{"id":"c3150c16-f3c2-4359-b597-1ddac2005e8b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:02:32.921479Z","strongest_claim":"CRePE represents each image token as a depth-aware positional distribution along its source ray, providing a Unified Camera Model-compatible positional encoding that captures the projected-path geometry induced by wide-angle and fisheye cameras.","one_line_summary":"CRePE supplies depth-aware positional distributions along curved rays for stable unified-camera control in frozen video DiT models.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That pseudo-supervision from a monocular geometry foundation model is sufficient to stabilize the Geometric Attention Adapter without introducing systematic bias in the learned ray distributions.","pith_extraction_headline":"CRePE represents each image token as a depth-aware positional distribution along its source ray to support unified camera control under the Unified Camera Model."},"references":{"count":24,"sample":[{"doi":"","year":2025,"title":"Recammaster: Camera-controlled generative rendering from a single video","work_id":"d26e2d77-d93f-42ea-83d3-eb5325fe30d4","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2026,"title":"arXiv preprint arXiv:2601.15275 (2026) 4, 8, 9, 21","work_id":"51636ad9-ac46-4382-bea0-d73080115ee4","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"arXiv preprint arXiv:2512.07237 (2025)","work_id":"59c944cc-e7b5-4565-84db-e5b36b9aeef5","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Scalable Diffusion Models with Transformers","work_id":"a3a05169-18b1-42bb-8775-eada50163437","ref_index":4,"cited_arxiv_id":"2212.09748","is_internal_anchor":true},{"doi":"","year":2025,"title":"Wan: Open and Advanced Large-Scale Video Generative Models","work_id":"ad3ebc3b-4224-46c9-b61d-bcf135da0a7c","ref_index":5,"cited_arxiv_id":"2503.20314","is_internal_anchor":true}],"resolved_work":24,"snapshot_sha256":"65d65b48e59bfd895c0057157d23a675fda24961fb87df36b2fe54d79f6df34e","internal_anchors":3},"formal_canon":{"evidence_count":2,"snapshot_sha256":"fa861051cd04929a8de81537a17ff8316cc369c0b46c6105fb0a58cd0d6310e8"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}