Research map / 3D vision / Gaussian splatting
DiffPBR: Point-Based Rendering via Spatial-Aware Residual Diffusion
Yiping Xie, Yuchi Huo, Yunlong Ran, Zijian Huang, Lincheng Li, Yingfeng Chen, Jiming Chen, Qi Ye
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 |
|---|---|---|---|
| use residual diffusion (RDDM) instead of DDPM with 2D random noise (ablation) | DiffPBR · ScanNet | PSNR: 19.22 -> 21.54 | Improved |
| add 3D consistent noise instead of 2D random noise (ablation) | RDDM · ScanNet | PSNR: 21.54 -> 22.15 | Improved |
| add adaptive global scale regularization on KNN scales (ablation) | DiffPBR-Q · ScanNet | PSNR after refinement: 22.05 -> 23.28 | Improved |
| add both coverage and compactness scale losses (ablation) | DiffPBR-Q · ScanNet | PSNR: 22.31 -> 23.28 | Improved |
| add coverage loss Lcov (ablation) | DiffPBR-Q · ScanNet | PSNR: 22.69 -> 23.28 | Improved |
| add compactness loss Lcmp (ablation) | DiffPBR-Q · ScanNet | PSNR: 22.97 -> 23.28 | Improved |
| replace KNN plus adaptive scale with pretrained CNN scale predictor (ablation) | DiffPBR-Q · ScanNet | PSNR after refinement: 23.28 -> 21.08 | Hurt |
| replace KNN plus adaptive scale with pretrained MLP scale predictor (ablation) | DiffPBR-Q · ScanNet | PSNR after refinement: 23.28 -> 21.47 | Hurt |
| use KNN scale calculation instead of MLP scale prediction (ablation) | adaptive splatting · ScanNet (100k points) | runtime ms: 525 -> 5.35 | Improved |
| add noise to point positions as well as color for noise splatting (ablation) | CoNo-Splatting · THuman2.0 | PSNR: 41.27 -> 40.02 | Hurt |
| use median instead of 25th percentile for global scale truncation (ablation) | DiffPBR-Q · ScanNet | PSNR: 22.41 -> 23.28 | Improved |
| use one-step x0 prediction instead of multi-step epsilon prediction (ablation) | DiffPBR · ScanNet | PSNR: 23.28 -> 22.92 | Hurt |
Closest papers by meaning
- DIFIX3D+: Improving 3D Reconstructions with Single-Step Diffusion Models (CVPR 2025 · similarity 0.90)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.90)
- pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction (CVPR 2024 · similarity 0.89)
- Evolving High-Quality Rendering and Reconstruction in a Unified Framework with Contribution-Adaptive Regularization (CVPR 2025 · similarity 0.89)
- SplatFormer: Point Transformer for Robust 3D Gaussian Splatting (ICLR 2025 · similarity 0.89)
- Neural Point Cloud Diffusion for Disentangled 3D Shape and Appearance Generation (CVPR 2024 · similarity 0.89)
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.
