Reproduce “Geometry beneath the Waves: Dense Priors for Sparse-View Underwater 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
Harvey Caldeira, Haoran Wang, Guoxi Huang, Shaoyu Cai, Rachel Fu, Nantheera Anantrasirichai
The authors adapt VGGT for underwater images and use predicted 3D geometry to generate pseudo-views for sparse-view reconstruction. Teacher–student training pairs degraded images with clean geometric supervision, and predicted geometry initializes the reconstruction pipeline. Across four SeaThru-NeRF scenes, the method reaches 27.11 dB PSNR, 0.8634 SSIM and 0.2188 LPIPS, exceeding COLMAP-initialized RUSplatting at 15k iterations; the comparison does not isolate pseudo-view generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.