Reproduce “Differentiable Voronoi Ray Tracing Beyond Rasterization Speeds” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bernardo Taveira, Carl Lindström, Joakim Johnander, Fredrik Kahl
VoroTracing renders novel views by tracing camera rays through neighboring cells in a 3D Voronoi partition. It reduces per-cell appearance costs with compact textures and concentrates opacity near surfaces to shorten traversal, with GPU optimizations. On seven Mip-NeRF 360 scenes, it averages 623 FPS, 28.98 dB PSNR, 0.848 SSIM, and 0.235 LPIPS on an RTX 5090, for pinhole-view benchmarks.
Key insights
- Differentiable ray tracing, when meticulously optimized through scene representation and GPU execution co-design, can surpass rasterization throughput in real-time novel view synthesis.
- A combination of octahedral textures, scale-invariant density, and surface-concentrated opacity via distortion loss significantly reduces ray traversal and per-cell work.
- Comprehensive inference-time optimizations, including Morton ordering, warp-coherent tiling, and low-contribution cell skipping, are crucial for achieving competitive real-time performance on GPUs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.