{"paper":{"title":"HarmoGS: Robust 3D Gaussian Splatting in the Wild via Conflict-Aware Gradient Harmonization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"HarmoGS resolves conflicting gradients in wild 3D Gaussian Splatting by refining masks with semantic consistency and rotating view gradients into orthogonal alignment.","cross_cats":[],"primary_cat":"cs.CV","authors_text":"Jian-Fang Hu, Jianhuang Lai, Tianze Zhu, Wei-Shi Zheng, Yulei Kang","submitted_at":"2026-05-13T06:47:09Z","abstract_excerpt":"In-the-wild 3D Gaussian Splatting remains challenging due to transient distractors and illumination-induced cross-view appearance inconsistencies. Existing methods mainly rely on image-level masking to suppress unreliable supervision, but masking alone cannot fully eliminate residual occlusions or resolve illumination-induced inconsistencies, both of which can introduce conflicting cross-view gradients. These unresolved conflicts may destabilize Gaussian optimization and lead to visible reconstruction artifacts. We propose a conflict-aware 3DGS framework that addresses this problem from both i"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"We propose a conflict-aware 3DGS framework that addresses this problem from both image-space supervision and gradient-level optimization. ... our method achieves state-of-the-art rendering quality under complex transient distractors and cross-view inconsistencies.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The assumption that semantic consistency-guided masking can accurately learn pixel-wise consistency scores to suppress unreliable supervision and that mutually rotating view-specific gradients into an orthogonal configuration reduces negative directional interference without losing useful optimization signal.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"HarmoGS improves in-the-wild 3D Gaussian Splatting by using semantic consistency-guided masking and dual-view conflict-aware gradient harmonization to reduce artifacts from transient distractors and cross-view inconsistencies.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"HarmoGS resolves conflicting gradients in wild 3D Gaussian Splatting by refining masks with semantic consistency and rotating view gradients into orthogonal alignment.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"fe98a6393df95403c1e4d2922ce7c289da8b18833a1a0b7f1891fea6967a4bb2"},"source":{"id":"2605.13073","kind":"arxiv","version":1},"verdict":{"id":"34e2dfc1-bc98-491c-9839-a5a98e4f268b","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:14:25.946078Z","strongest_claim":"We propose a conflict-aware 3DGS framework that addresses this problem from both image-space supervision and gradient-level optimization. ... our method achieves state-of-the-art rendering quality under complex transient distractors and cross-view inconsistencies.","one_line_summary":"HarmoGS improves in-the-wild 3D Gaussian Splatting by using semantic consistency-guided masking and dual-view conflict-aware gradient harmonization to reduce artifacts from transient distractors and cross-view inconsistencies.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The assumption that semantic consistency-guided masking can accurately learn pixel-wise consistency scores to suppress unreliable supervision and that mutually rotating view-specific gradients into an orthogonal configuration reduces negative directional interference without losing useful optimization signal.","pith_extraction_headline":"HarmoGS resolves conflicting gradients in wild 3D Gaussian Splatting by refining masks with semantic consistency and rotating view gradients into orthogonal alignment."},"references":{"count":37,"sample":[{"doi":"","year":2022,"title":"Tensorf: Tensorial radiance fields","work_id":"dd4b09a1-6330-4fbf-8f26-0f12f1a94414","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Nerf-hugs: Improved neural radiance fields in non-static scenes using heuristics-guided segmentation","work_id":"9078c431-4250-4b59-9690-3390a557e7ed","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Hallucinated neural radiance fields in the wild","work_id":"a0f49056-909e-4ea7-839c-150f414433a0","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"K-planes: Explicit radiance fields in space, time, and appearance","work_id":"30dcc940-6c9a-457f-94c3-6e2634f6c343","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2022,"title":"Plenoxels: Radiance fields without neural networks","work_id":"31d38ed8-2831-43d2-9b63-ceb9550be5aa","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":37,"snapshot_sha256":"ffac81dc6250539fd0646a64a22fc8b0e0c332064fd7096ccb3bb1b5daeab964","internal_anchors":1},"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"}