Reproduce “LGS: A Light-weight 4D Gaussian Splatting for Efficient Surgical Scene Reconstruction” 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

Hengyu Liu, Yifan Liu, Chenxin Li, Wuyang Li, Yixuan Yuan

The LGS framework introduces a lightweight 4D Gaussian Splatting approach for dynamic surgical scene reconstruction, significantly reducing model memory footprint by 9x to 15x while maintaining real-time rendering speeds exceeding 180 FPS and high visual fidelity. This enables practical deployment on resource-limited surgical equipment, making advanced 4D scene understanding feasible for clinical applications.

Key insights

  • Surgical scenes often have simpler textures and dynamics than natural scenes, allowing for significant compression of Gaussian attributes (SH coefficients) without substantial loss of visual quality.
  • Not all Gaussians contribute equally to representing dynamic deformations; identifying and pruning less important Gaussians can reduce redundancy.
  • Spatiotemporal feature fields, while crucial for dynamics, contain redundancy where nearby Gaussians share similar information, making resolution reduction via adaptive pooling effective.

Summary by alphaXiv.

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