Research map / 3D vision / Gaussian splatting
CoherentGS: Sparse Novel View Synthesis with Coherent 3D Gaussians
Avinash Paliwal, Wei Ye, Jinhui Xiong, Dmytro Kotovenko, Rakesh Ranjan, Vikas Chandra, Nima Khademi Kalantari
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 |
|---|---|---|---|
| add implicit convolutional decoder for per-view residual depth (ablation) | CoherentGS · LLFF (3 views) | PSNR: 16.68 -> 20.33 | Improved |
| add flow-based coarse alignment of monocular depth at initialization (ablation) | CoherentGS · LLFF (3 views) | PSNR: 19.06 -> 20.33 | Improved |
| add multisampling within each pixel (ablation) | CoherentGS · LLFF (3 views) | PSNR: 19.99 -> 20.33 | Improved |
| add TV regularization on rendered disparity (ablation) | CoherentGS · LLFF (3 views) | PSNR: 20.20 -> 20.33 | Improved |
| add flow-based regularization on Gaussian positions (ablation) | CoherentGS · LLFF (3 views) | PSNR: 20.32 -> 20.33 | Improved |
| add depth-based segmentation mask with 5 channels to decoder output (ablation) | CoherentGS · LLFF (3 views) | PSNR: 19.70 -> 20.33 | Improved |
| increase segmentation mask channels from 5 to 7 (ablation) | CoherentGS · LLFF (3 views) | PSNR: 20.33 -> 20.13 | Hurt |
| constrain Gaussians to per-pixel rays with decoder, versus unstructured 3DGS | 3DGS · LLFF (3 views) | PSNR: 14.99 -> 20.33 | Improved |
| CoherentGS versus SparseNeRF at 4 input views | CoherentGS · LLFF (4 views) | PSNR: 21.90 -> 21.58 | Hurt |
| CoherentGS versus SparseNeRF at 3 views on NVS-RGBD ZED 2 | CoherentGS · NVS-RGBD ZED 2 (3 views) | PSNR: 26.56 -> 24.93 | Hurt |
| 3DGS-based rendering instead of NeRF baselines | CoherentGS · LLFF (3 views) | inference speed: 0.08 fps -> 278 fps | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 3 on most shared results.
- SparseNeRF — SparseNeRF: Distilling Depth Ranking for Few-shot Novel View Synthesis (ICCV 2023) · better on 2 of 5 shared results
- FreeNeRF — FreeNeRF: Improving Few-Shot Neural Rendering With Free Frequency Regularization (CVPR 2023) · better on 2 of 3 shared results
- DNGaussian — DNGaussian: Optimizing Sparse-View 3D Gaussian Radiance Fields with Global-Local Depth Normalization (CVPR 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- FewViewGS: Gaussian Splatting with Few View Matching and Multi-stage Training (NeurIPS 2024 · similarity 0.93)
- DropGaussian: Structural Regularization for Sparse-view Gaussian Splatting (CVPR 2025 · similarity 0.92)
- Binocular-Guided 3D Gaussian Splatting with View Consistency for Sparse View Synthesis (NeurIPS 2024 · similarity 0.92)
- Compact 3D Gaussian Representation for Radiance Field (CVPR 2024 · similarity 0.92)
- COLMAP-Free 3D Gaussian Splatting (CVPR 2024 · similarity 0.92)
- NexusGS: Sparse View Synthesis with Epipolar Depth Priors in 3D Gaussian Splatting (CVPR 2025 · similarity 0.91)
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.
