Research map / 3D vision / Gaussian splatting
CL-Splats: Continual Learning of Gaussian Splatting with Local Optimization
Jan Ackermann, Jonas Kulhanek, Shengqu Cai, Haofei Xu, Marc Pollefeys, Gordon Wetzstein, Leonidas J. Guibas, Songyou Peng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| freeze Gaussians outside changed region (ablation) | CL-Splats · CL-Splats synthetic Level-2 | PSNR: 20.773 -> 40.833 | Improved |
| use majority voting instead of requiring all-view projection (ablation) | CL-Splats · CL-Splats synthetic Level-2 | PSNR: 35.611 -> 40.833 | Improved |
| add local optimization kernel (ablation) | CL-Splats · CL-Splats synthetic Level-2 | optimization time: 8 min -> 5 min | Improved |
| add local optimization kernel (ablation) | CL-Splats · CL-Splats synthetic Level-2 | PSNR: 40.812 -> 40.833 | No clear effect |
| use spheres instead of rectangle primitives (ablation) | CL-Splats · CL-Splats synthetic Level-2 | PSNR: 40.717 -> 40.833 | No clear effect |
| lift 2D change masks to 3D via local optimization (ablation) | CL-Splats · mask quality | precision: 0.370 -> 0.609 | Improved |
| add mask dilation (ablation) | CL-Splats · mask quality | precision: 0.611 -> 0.609 | No clear effect |
| use DINOv2 change mask instead of color L2 baseline (ablation) | CL-Splats · mask quality | recall: 0.761 -> 0.961 | Improved |
| restrict 3DGS photometric loss to 2D masks (3DGS+M) instead of vanilla 3DGS | 3DGS · CL-Splats synthetic | PSNR: 21.993 -> 15.127 | Hurt |
| use CL-Splats instead of CL-NeRF | CL-NeRF · CL-Splats synthetic | PSNR: 30.063 -> 40.125 | Improved |
| use CL-Splats instead of CL-NeRF | CL-NeRF · CL-NeRF dataset | PSNR: 27.302 -> 29.984 | Improved |
| incremental pose registration instead of joint COLMAP (ablation) | CL-Splats · real-world scenes | PSNR: average PSNR drop of 1.2, up to 20 times speedup | Hurt |
Closest papers by meaning
- MCGS: Markov Chain Gaussian Splatting for Dynamic Scenes Reconstruction (AAAI 2026 · similarity 0.90)
- Learning Efficient Fuse-and-Refine for Feed-Forward 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.89)
- SC-GS: Sparse-Controlled Gaussian Splatting for Editable Dynamic Scenes (CVPR 2024 · similarity 0.88)
- Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstruction (CVPR 2024 · similarity 0.88)
- Segment then Splat: Unified 3D Open-Vocabulary Segmentation via Gaussian Splatting (NeurIPS 2025 · similarity 0.88)
- DGS-LRM: Real-Time Deformable 3D Gaussian Reconstruction From Monocular Videos (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
