Reproduce “DeltaSplat: Iterative Gaussian Refinement for Pose-Free Feed-Forward 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

Chanung Park, Seunghyeon Song, Joo Chan Lee, Eunbyung Park, Jong Hwan Ko

DeltaSplat refines feed-forward 3D Gaussian Splatting scenes reconstructed from a few unposed images. It renders the current scene at input views, then uses image residuals, per-pixel Plücker rays, and rendered depth to update Gaussian position, opacity, and color over three iterations. On DL3DV, it reaches 26.64 dB PSNR, 1.75 dB above its YoNoSplat backbone, with known intrinsics and predicted poses.

Summary by alphaXiv.

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