Accelsnow / diff-gaussian-rasterization-distwar

DISTWAR-enabled rasterization engine for the paper "3D Gaussian Splatting for Real-Time Rendering of Radiance Fields"

DISTWAR atomic reduction on Differential Gaussian Rasterization

Modified 3DGS rasterization engine for the paper "3D Gaussian Splatting for Real-Time Rendering of Radiance Fields" with DISTWAR(paper here) atomic reduction optimizations. We apply two different DISTWAR optimizations to the backward kernel for gaussian rasterization - serialized atomic reduction SW-S and butterfly atomic reduction SW-B. The modified backward kernels are implemented in cuda_rasterizer/backward.cu.

This is a submodule of a modified 3DGS repository. For detailed instruction on how to enable different DISTWAR optimization modes, please refer to the parent repository.

Default configuration: BW_IMPLEMENTATION=0   BALANCE_THRESHOLD=8

Citation for original 3DGS paper:

BibTeX

@Article{kerbl3Dgaussians,
      author       = {Kerbl, Bernhard and Kopanas, Georgios and Leimk{\"u}hler, Thomas and Drettakis, George},
      title        = {3D Gaussian Splatting for Real-Time Radiance Field Rendering},
      journal      = {ACM Transactions on Graphics},
      number       = {4},
      volume       = {42},
      month        = {July},
      year         = {2023},
      url          = {https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/}
}