Reproduce “Endo-4DGS: Endoscopic Monocular Scene Reconstruction with 4D 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
Yiming Huang, Beilei Cui, Long Bai, Ziqi Guo, Mengya Xu, Mobarakol Islam, Hongliang Ren
Endo-4DGS introduces a real-time endoscopic monocular dynamic scene reconstruction method using a 4D Gaussian Splatting representation. It generates high-fidelity 3D models of deformable tissues at 100 FPS by leveraging pseudo-depth priors and confidence-guided learning, significantly improving speed and reducing computational resources compared to previous approaches.
Key insights
- Extending 3D Gaussian Splatting to 4D with a dedicated deformation network allows for an efficient and high-fidelity representation of dynamic and deformable endoscopic surgical scenes.
- Pseudo-depth priors, even from monocular estimation models, can effectively initialize and guide 4D Gaussian Splatting optimization when combined with confidence weighting and geometric constraints.
- The synergistic application of confidence-guided learning, depth regularization, and surface normal constraints is crucial for achieving geometrically consistent and visually accurate reconstructions from inherently noisy monocular depth estimates.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.