Reproduce “Balanced 3DGS: Gaussian-wise Parallelism Rendering with Fine-Grained Tiling” 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
Hao Gui, Lin Hu, Rui Chen, Mingxiao Huang, Yuxin Yin, Jin Yang, Yong Wu, Chen Liu and 3 more
Balanced 3DGS redistributes rendering work during 3D Gaussian Splatting training without changing the scene representation. Warps cooperate across Gaussian contributions, while blocks dynamically take smaller image tiles; the system switches to a memory-optimized kernel when balancing no longer helps. In the tested setup, it reports 7.52× faster forward-render kernel time and 8.5% faster end-to-end training overall.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.