Reproduce “RoSplat: Robust Feed-Forward Pixel-wise Gaussian Splatting for Varying Input Views and High-Resolution Rendering” 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
Hoang Chuong Nguyen, Renjie Wu, Jose M. Alvarez, Miaomiao Liu
RoSplat targets novel-view synthesis by reweighting overlapping Gaussians using depth consistency and adding training-only supervision for Gaussian scales. Its scale branch is removed at inference. On RealEstate10K, alpha normalization raises DepthSplat’s 16-view PSNR from 20.97 to 25.86 dB without retraining. On DL3DV, the regularizer raises PSNR from 18.11 to 19.38 dB at 2048 × 3584, with gains in these tested settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.