Reproduce “ZipSplat: Fewer Gaussians, Better Splats” 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
Alexander Veicht, Sunghwan Hong, Dániel Baráth, Marc Pollefeys
ZipSplat reconstructs 3D scenes from unposed images using tokens initialized from image patches, merging observations at an inference-time budget while retaining patch-to-3D associations for geometric refinement. On DL3DV with 16 input views, it achieves 24.79 dB PSNR with 166K Gaussians, versus TokenGS’s 21.41 dB with 262K; this comparison uses different camera conditions and does not isolate components.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.