Reproduce “FreeTimeGS++: Secrets of Dynamic Gaussian Splatting and Their Principles” 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
Lucas Yunkyu Lee, Soonho Kim, Youngwook Kim, Sangmin Kim, Jaesik Park
FreeTimeGS++ analyzes and extends FreeTimeGS for reconstructing dynamic scenes with 4D Gaussian Splatting, formalizing design choices and adding flow-guided initialization, color correction, and a persistence gate. On six-scene DyNeRF aggregates, it raises PSNR from 32.62 ± 0.14 to 33.40 ± 0.09 across six runs. The study centers on FreeTimeGS and does not establish motion accuracy.
Key insights
- Temporal lifetimes emerge as an implicit division of labor: long-lived Gaussians mainly model backgrounds, while short-lived primitives tend to capture dynamic subjects, even without explicit static/dynamic labels.
- Rendering accuracy and motion consistency are partly decoupled; noisy velocities can still produce visually accurate frames because the image loss does not require persistent primitive identities.
- Initialization should match both scene geometry and motion complexity, while density relocation and color correction influence not only PSNR but also artifacts and run-to-run reliability.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.