Reproduce “HOMER: Homography-Based Efficient Multi-view 3D Object Removal” 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

Jingcheng Ni, Weiguang Zhao, Daniel Wang, Ziyao Zeng, Chenyu You, Alex Wong, Kaizhu Huang

HOMER is a framework for multi-view 3D object removal that reduces user interaction and total removal time by using homography-based mask propagation and selective inpainting. It demonstrates compatibility with both NeRF and 3D Gaussian Splatting, achieving an 80% reduction in removal time compared to prior methods.

Key insights

  • Homography-based transformations can efficiently and robustly propagate 2D masks and inpainted content across multiple views in a 3D scene, significantly reducing computational overhead.
  • Leveraging readily available 2D segmentation and inpainting models within a multi-view pipeline simplifies user interaction and streamlines initial content generation.
  • Adaptive refinement mechanisms, such as the anchor circle adjustment for masks and selective full inpainting on key views, effectively correct accumulated errors from warping and maintain multi-view consistency.

Summary by alphaXiv.

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