Reproduce “Dynamic Gaussian Marbles for Novel View Synthesis of Casual Monocular Videos” 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

Colton Stearns, Adam Harley, Mikaela Uy, Florian Dubost, Federico Tombari, Gordon Wetzstein, Leonidas Guibas

Gaussian Marbles reconstructs dynamic scenes from single-camera video for novel-view rendering and tracking. It uses spherical Gaussians, local motion estimation, and global adjustment, guided by image tracks and geometry and rendering cues. In DyCheck’s seven-scene no-camera-pose setting, it reports mean mPSNR 15.79 and LPIPS 0.428, versus 13.11 and 0.726 for 4D Gaussians; DyCheck’s camera path still gathers multiview information.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.