Reproduce “Relaxed Rigidity with Ray-based Grouping for Dynamic 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
Junoh Lee, Junmyeong Lee, Yeon-Ji Song, Inhwan Bae, Jisu Shin, Hae-Gon Jeon, Jin-Hwa Kim
Researchers at GIST, Yonsei Univ., SNU, DGIST, and NAVER AI Lab developed a framework for dynamic 3D Gaussian Splatting that enforces relaxed rigidity constraints through ray-based grouping and novel regularization. This approach consistently enhanced reconstruction quality and temporal consistency across diverse dynamic scene datasets and baselines.
Key insights
- Utilizing the 3DGS rasterization process for ray-based grouping provides an efficient, adaptive, and occlusion-aware method to form physically coherent Gaussian groups, which is superior to distance-based grouping.
- Relaxed rigidity constraints, achieved through motion coherence and spectral regularization, are more effective than strict rigidity assumptions for accurately modeling diverse non-rigid deformations in dynamic scenes.
- Combining these regularization techniques allows dynamic 3DGS models to learn physically plausible motion directly from image supervision, thereby reducing reliance on potentially inconsistent external priors.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.