Reproduce “POTR: Post-Training 3DGS Compression” 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
Bert Ramlot, Martijn Courteaux, Peter Lambert, Glenn Van Wallendael
POTR compresses already-trained 3D Gaussian Splatting scenes by estimating the image-quality cost of removing each splat, then refitting view-dependent spherical-harmonics coefficients for cheaper encoding. In Garden, at PSNR 27, it requires 29% fewer splats than MesonGS and LightGaussian. Across four datasets, it reports strong rate-distortion results without fine-tuning, but encoding takes 20–170 seconds per model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.