Reproduce “TRiGS: Temporal Rigid-Body Motion for Scalable 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
Suwoong Yeom, Joonsik Nam, Seunggyu Choi, Lucas Yunkyu Lee, Sangmin Kim, Jaesik Park, Joonsoo Kim, Kugjin Yun and 2 more
TRiGS introduces a 4D Gaussian Splatting framework that leverages coupled SE(3) rigid-body motion with hierarchical Bézier residuals and local anchors to reconstruct dynamic scenes. This method scales effectively to extended video sequences up to 1200 frames, maintaining a constant memory footprint of 0.5M Gaussians and 160MB while achieving high rendering quality and over 110 FPS.
Key insights
- Parameterizing rigid-body motion directly in the Lie algebra se(3) provides a stable and coupled transformation, outperforming uncoupled rotation and translation optimizations.
- Decomposing motion into a base term and hierarchical Bézier residuals allows capturing complex, non-linear trajectories while maintaining primitive identity over longer durations.
- Employing learnable local anchors, coupled with gauge fixing, enables precise localized deformations and resolves optimization ambiguities in identifying translational components.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.