Neural denoiser speeds up noise removal in new 3d gaussian rendering
Ultra-fast Neural Inference for Stochastic Gaussian Splatting Denoising
Computer Vision and Pattern RecognitionGraphics
Summary
Rendering 3D images using Gaussian splatting can be slow because it requires sorting and blending many visual elements. The authors found a way to skip these slow steps by allowing some noise in the image and then using a special neural network to quickly clean up the noise over time. This method produces stable and clear images even when the viewpoint changes, while still running faster overall. Their approach keeps the final image quality close to the slower method, but with better performance.
What this means in practice
- •For graphics programmers: Create faster 3D rendering pipelines by reducing sorting and blending while maintaining image stability and quality.
- •For game engine developers: Improve real-time visual quality during camera movement using neural denoising to speed up stochastic Gaussian splatting.
Authors
Chenxiao Hu, Hao Zhang, Yanchen Zhang, Meng Gai, Guoping Wang, Sheng Li
Abstract
Stochastic rendering eliminates the sorting and alpha blending process in Gaussian splatting, at the cost of introducing spatial noise. Formulating temporal denoising over the pixel stream shared by view-consistent stochastic splatting renderers, we propose a temporal neural denoiser validated on stochastic 2D Gaussian Splatting rendering, combining dual-path exponential moving average accumulation, per-pixel learned trust prediction for history validation, a fixed anisotropic spatial filter and a variance-gated composition with stabilization. The denoiser suppresses the noise, achieving temporally stable, visually compelling outputs during free camera navigation, all while retaining the sort-free, blend-free rasterization performance. The combined pipeline retains a PSNR gap to sorted alpha-blending renderers, but the denoiser's overhead stays below the time saved by removing sorting and blending.