Interactive 3D Gaussian Splatting & Differentiable Rendering Laboratory
An educational platform that teaches the theory and practice of 3D Gaussian Splatting (3DGS) — the breakthrough technique that replaced NeRFs for real-time radiance field rendering. Zero dependencies, single HTML file.
-
Gaussian Primitives Explorer — Manipulate 3D Gaussian parameters (position, rotation, scale, color, opacity) and watch real-time projection from 3D ellipsoid to 2D splat via
Σ' = J·Σ·J⊤ -
Splatting Pipeline Visualizer — Step through the complete forward rendering pipeline: world→camera transform, 2D covariance projection, depth sorting, tile assignment, and alpha compositing
-
Differentiable Rasterizer — The key innovation: click any pixel to see how gradients flow backward through the rendering pipeline to each Gaussian's parameters (∂L/∂μ, ∂L/∂Σ, ∂L/∂α, ∂L/∂c)
-
Scene Optimizer — Watch 120+ random Gaussians optimize in real-time to reconstruct target images via gradient descent on L2 pixel loss
-
Adaptive Density Control — Interactive demonstration of splitting (large + high gradient), cloning (small + high gradient), and pruning (low opacity) operations
-
3DGS vs NeRF Arena — Side-by-side race: Gaussian splatting vs MLP-based neural radiance field, with live PSNR tracking and comparison metrics
- Single HTML file (~85KB)
- Zero external dependencies
- Canvas2D rendering with custom math library
- Real-time gradient computation
- Client-side optimization
- Kerbl et al., "3D Gaussian Splatting for Real-Time Radiance Field Rendering", SIGGRAPH 2023
- Mildenhall et al., "NeRF: Representing Scenes as Neural Radiance Fields", ECCV 2020
- Huang et al., "2D Gaussian Splatting for Geometrically Accurate Radiance Fields", SIGGRAPH 2024
- Triangle Splatting (arXiv 2505.19175, May 2026)
MIT