{"paper":{"title":"RoSplat: Robust Feed-Forward Pixel-wise Gaussian Splatting for Varying Input Views and High-Resolution Rendering","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"Alpha normalization and a 3D regularizer make pixel-wise Gaussian splatting produce consistent brightness and fewer holes regardless of input view count or output resolution.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Hoang Chuong Nguyen, Jose M. Alvarez, Miaomiao Liu, Renjie Wu","submitted_at":"2026-05-13T07:05:15Z","abstract_excerpt":"Generalizable 3D Gaussian Splatting has recently emerged as an efficient approach for novel-view synthesis, enabling feed-forward synthesis from only a few input views. However, existing pixel-wise feed-forward methods suffer from over-bright renderings when the number of input views varies during inference, as well as insufficient supervision for accurate Gaussian scale estimation, which leads to hole artifacts, particularly in high-resolution renderings. To address these issues, we identify that the over-brightness is caused by the varying number of overlapping Gaussians and propose a simple"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"we identify that the over-brightness is caused by the varying number of overlapping Gaussians and propose a simple alpha normalization strategy to maintain brightness consistency across different number of input views. In addition, we introduce an auxiliary 3D sampling-based regularizer to improve Gaussian scale estimation, thereby mitigating hole artifacts in high-resolution rendering.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that over-brightness stems solely from varying numbers of overlapping Gaussians and that the proposed alpha normalization plus 3D regularizer will generalize without introducing new artifacts on unseen scenes or data distributions.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"RoSplat adds alpha normalization for brightness consistency across varying input views and a 3D sampling regularizer to mitigate hole artifacts in high-resolution feed-forward Gaussian splatting.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Alpha normalization and a 3D regularizer make pixel-wise Gaussian splatting produce consistent brightness and fewer holes regardless of input view count or output resolution.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b0be212524eb9fdbf0ec40add8476ee15acc1bc05b959d2b2bf30304048a3f9a"},"source":{"id":"2605.13093","kind":"arxiv","version":1},"verdict":{"id":"a36c6360-cc01-45f1-b135-5e4d6df62222","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T19:58:12.242402Z","strongest_claim":"we identify that the over-brightness is caused by the varying number of overlapping Gaussians and propose a simple alpha normalization strategy to maintain brightness consistency across different number of input views. In addition, we introduce an auxiliary 3D sampling-based regularizer to improve Gaussian scale estimation, thereby mitigating hole artifacts in high-resolution rendering.","one_line_summary":"RoSplat adds alpha normalization for brightness consistency across varying input views and a 3D sampling regularizer to mitigate hole artifacts in high-resolution feed-forward Gaussian splatting.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that over-brightness stems solely from varying numbers of overlapping Gaussians and that the proposed alpha normalization plus 3D regularizer will generalize without introducing new artifacts on unseen scenes or data distributions.","pith_extraction_headline":"Alpha normalization and a 3D regularizer make pixel-wise Gaussian splatting produce consistent brightness and fewer holes regardless of input view count or output resolution."},"references":{"count":30,"sample":[{"doi":"","year":2021,"title":"Mip-nerf: A multiscale representation for anti-aliasing neural radiance fields","work_id":"06181443-2ada-4ffb-8f93-dbe50d117b8f","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Zip-nerf: Anti- aliased grid-based neural radiance fields","work_id":"50cc7302-9c45-4628-bb11-faa781c35ea7","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Textured gaussians for enhanced 3d scene appearance modeling","work_id":"0fec5157-cbc4-427b-8c52-bedfa56e21eb","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"pixelsplat: 3d gaussian splats from image pairs for scalable generalizable 3d reconstruction","work_id":"9f63c70b-67ae-46ef-812f-4660a72d3e56","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Mvsnerf: Fast generalizable radiance field reconstruction from multi-view stereo","work_id":"1f81c980-9a93-4ccb-9bbd-c7a269b7c44b","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":30,"snapshot_sha256":"3edd025d38fb5ec651edc51ee806293e3ae7abec90885a896352ecdbce2508a7","internal_anchors":1},"formal_canon":{"evidence_count":2,"snapshot_sha256":"ff1e5f09fa31ff1443a1641c1c5b9cc196369e22fd3c8ea952b8529a088439ab"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}