Reproduce “DWTGS: Rethinking Frequency Regularization for Sparse-view 3D Gaussian Splatting” 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
Hung Nguyen, Runfa Li, An Le, Truong Nguyen
DWTGS improves sparse-view 3D Gaussian Splatting by replacing fully supervised frequency matching with multi-scale wavelet losses. It matches low-frequency LL bands in training views and encourages sparse first-level HH detail at novel viewpoints without ground-truth targets. On held-out 3-view LLFF, it reaches 20.78 PSNR, 0.720 SSIM, and 0.197 LPIPS, versus FreGS’s 20.40, 0.706, and 0.198. The comparison covers tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.