Reproduce “Intrinsic Geometry-Appearance Consistency Optimization for Sparse-View 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

Kaiqiang Xiong, Rui Peng, Jiahao Wu, Zhanke Wang, Jie Liang, Xiaoyun Zheng, Feng Gao, Ronggang Wang

A framework called ICO-GS enhances 3D Gaussian Splatting performance in sparse-view settings by enforcing intrinsic geometry-appearance consistency through robust geometric regularization and geometry-guided appearance optimization. The method achieves state-of-the-art results, improving PSNR by up to +1.06 dB on the DTU 3-view dataset and +0.76 dB on the LLFF 3-view dataset over existing baselines.

Key insights

  • Intrinsic geometry-appearance consistency, where correct geometry and coherent appearance mutually reinforce each other, is crucial for faithful 3D reconstruction, especially in sparse-view conditions.
  • Standard sparse-view 3DGS optimization suffers from insufficient geometric constraints and unreliable appearance-geometry coupling due to "appearance compensation."
  • Robust, internal multi-view geometric regularization combined with appearance optimization guided by geometrically validated regions can overcome these limitations without solely relying on potentially noisy external priors.

Summary by alphaXiv.

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