A pure pytorch implementation of 3D gaussian splatting.
clone the repo
git clone https://github.com/hbb1/torch-splatting.git --recursive
and run
python train.py
Tile-based rendering is implemeted. Because running loop for python is slow, it uses 64x64-sized tile instead of 16x16 as 3DGSS did. The training time is about 2 hours for 512x512 resolution image for 30k iterartions, tested on a RTX 2080Ti. The number of 3D gaussians is fixed, of 16384 points. Under this setting, it matchs the original diff-gaussian-splatting implementation (~39 PSNR on my synthetic data).
Stay Tuned.
https://github.com/graphdeco-inria/gaussian-splatting/tree/main
https://github.com/graphdeco-inria/diff-gaussian-rasterization