Reproduce “SplatFill: 3D Scene Inpainting via Depth-Guided 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
Mahtab Dahaghin, Milind G. Padalkar, Matteo Toso, Alessio Del Bue
SplatFill removes objects from editable 3D scenes by first creating a diffusion-based 2D inpainted reference, organizing 3D Gaussian primitives with image, depth, and object-aware supervision, then selectively refining views and regions with depth disagreement. On 10 SPIn-NeRF scenes, training took 34 versus 45 minutes for GScream, while full-image FID was 29.76 versus 36.72; masked-region scores were tied.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.