Reproduce “CoherentGS: Sparse Novel View Synthesis with Coherent 3D Gaussians” 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
Avinash Paliwal, Wei Ye, Jinhui Xiong, Dmytro Kotovenko, Rakesh Ranjan, Vikas Chandra, Nima Khademi Kalantari
CoherentGS reconstructs navigable 3D scenes from two to four posed photographs by assigning Gaussians to input pixels, restricting positions to camera rays, and coupling depth updates with cross-view matches and smooth residuals. On LLFF’s three-view benchmark, it matches SparseNeRF at PSNR 20.33, while improving SSIM and LPIPS; at four views, its PSNR is lower, and hidden regions remain unreconstructed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.