Make sure you have installed Nvidia driver (version > 530), CUDA (version > 12) and Docker. It may be a lot easier to just use Windows with WSL, as there is an existing Pre-built Windows Binaries for viewer so you don't have to build it yourself. But you can still follow this if you are using ubuntu 22.04 (or anyother docker compatible linux distro) but you will have to either vuild that viewer on Linux or mount the trained model to a windows machine to use prebuilt viewer
Either pull gaetanlandreau/3d-gaussian-splatting from you GUI docker app in Windows or run the following command in your terminal:
docker pull gaetanlandreau/3d-gaussian-splatting:latest
Once you have this image pulled, clone 3dgs code into your current working directory:
git clone --recursive https://github.com/graphdeco-inria/gaussian-splatting.git
Start your docker in nvidia-runtime mode with their code mounted in container:
docker run --rm --runtime=nvidia --gpus all -v ./gaussian-splatting:/root/gaussian-splatting -it sha256:ac7c0eedae703a2d8c1b366152d766b57c924887bfffdcc178c676162d580dc4 bash
You will need to update this docker's "diff-gaussian-rasterization" "simple-knn" python package, those are submodules of the original 3dgs project.
cd /root/gaussian-splatting
pip install -q ./submodules/diff-gaussian-rasterization
pip install -q ./submodules/simple-knn
Once those packages are installed, download the data and train the model, it took ~15 mins with my RTX 3080ti
wget https://huggingface.co/camenduru/gaussian-splatting/resolve/main/tandt_db.zip
unzip tandt_db.zip
python train.py -s ./tandt/train
To see your rendering, inside windows, download and decompress their viewer In powershell, cd into PATH_TO_YOUR_DOWNLOAD_FOLDER\viewers\bin and run something like:
.\SIBR_gaussianViewer_app.exe -m \\wsl.localhost\Ubuntu-22.04\home\<YOUR_USER_NAME>\gaussian-splatting\output\<run-id>
Render the view from training data'position and orentation from your trained gaussians
python render.py -m output/<run-id>/