Reproduce “Smaller and Faster 3DGS via Post-Training Dictionary Learning” 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
Jiarong Gong, Jonas Unger, Ehsan Miandji
The method compresses trained 3D Gaussian Splatting models by replacing much of their spherical-harmonic color coefficients with a shared learned dictionary and sparse codes, without retraining. On 13 scenes, it reports compression ratios of 3.95×, 3.10×, and 4.55× for 3DGS, MCMC, and PixelGS, with FPS gains of 23.3%, 24.3%, and 25.3%, respectively; these are SH-data and rendering-throughput results, not deployment tests on constrained devices.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.