Reproduce “Monocular Dynamic Gaussian Splatting: Fast, Brittle, and Scene Complexity Rules” 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
Yiqing Liang, Mikhail Okunev, Mikaela Angelina Uy, Runfeng Li, Leonidas Guibas, James Tompkin, Adam W. Harley
This work conducts a systematic benchmark and critical analysis of monocular dynamic Gaussian Splatting methods, revealing that while these techniques offer rapid rendering, they exhibit brittleness and that scene characteristics heavily influence their reconstruction quality. The study identifies specific performance trade-offs among different motion representations and highlights challenges related to adaptive density control and reliance on multi-view cues.
Key insights
- Monocular dynamic Gaussian Splatting methods are faster than NeRF-like hybrid neural fields (TiNeuVox) but consistently achieve lower image quality and can be more brittle during optimization.
- Dataset characteristics, such as the complexity of motion and the availability of multi-view cues, often dominate over methodological differences in determining a method's performance.
- Adaptive density control, a core mechanism of 3DGS, introduces significant challenges in dynamic monocular settings, leading to efficiency variability, overfitting risks, and unpredictable optimization failures.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.