Reproduce “MCSplat: Multi-View Photometric and Geometric Consistent Feed-Forward Gaussian Splatting for Driving Scenes” 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

Yifei Wang, Wei Tian, Wenhua Bao, Jia Zhou, Hanshi Wang, Fudong Ge, Jianhua Wu, Shuai Yuan and 5 more

MCSplat targets inconsistent vehicle-camera appearance in feed-forward Gaussian splatting for driving scenes. It canonicalizes appearance before learning shared geometry, restores each camera’s look with bilateral-grid transforms, and weights geometric self-supervision by image-evidence uncertainty. On Waymo, it reduces adjacent-camera overlap mean depth reprojection error from 2.378 m to 0.910 m versus UniSplat, while normal references are pseudo-normals rather than ground truth.

Summary by alphaXiv.

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