Gaussian Splatting Alternative: WebGL Implementation of Nvidia's SVRaster

Summary

SVRaster WebGL Viewer A WebGL-based viewer for visualizing sparse voxel scenes from the Nvidia Sparse Voxels Rasterization paper. This viewer provides an interactive way to explore and visualize the voxel radiance field from the web. You can try the viewer at vid2scene.com/voxel The rendering isn't exactly the same as the reference CUDA implementation, but it's pretty similar. Features Interactive camera controls: Left-click + drag: Orbit camera Right-click + drag: Pan camera Mouse wheel: Zoom WASD/Arrow keys: Move camera Q/E: Rotate scene around view direction Space/Shift: Move up/down Touch controls for mobile devices: 1 finger drag: Orbit 2 finger drag: Pan/zoom Performance metrics display (FPS counter) How to Run Prerequisites Before running the project, you need to have Node.js and NPM (Node Package Manager) installed: Install Node.js and NPM: Download and install from nodejs.org Verify installation: node --version npm --version Running the Project To run this project locally: Clone the repository: git clone https://github.com/samuelm2/svraster-webgl.git cd svraster-webgl Install dependencies: npm install Start the development server: npm run dev This will start the Vite development server, typically at http://localhost:5173 Implementation and Performance Notes This viewer uses a distance-based sorting approach rather than the ray direction-dependent Morton ordering described in the paper The current implementation has only the most basic optimizations applied - there's significant room for performance improvements. Right now, the fragment shader is the bottleneck. Memory usage could also be lowered because nothing is quantized right now. If you have a perf improvement suggestion, please feel free to submit a PR! It runs at ~60-80 FPS on my laptop with a Laptop 3080 GPU It runs at about ~12-20 FPS on my iPhone 13 Pro Max Right now, only scenes trained with spherical harmonic degree 1 are supported (so 12 total SH coefficients per voxel). See the command below t...