Reproduce “WINGS: Reference-Free Gaussian Splatting Inpainting with 3D-Native Generative Priors” 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
Noé Lallouet, Michael Fischer, Elie Michel
The paper presents WINGS, which completes holes left by object removal directly in a learned 3D representation rather than using 2D-inpainted reference views. It predicts missing structure and features, then merges Gaussians with the scene. Across 29 scenes, WINGS has lower FID and Local-FID than Inpaint360GS, plus a 75.11% tie-adjusted win rate; results use predefined masks and take three minutes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.