• Clone this repo recursively: git clone git@ChikaYan/Gaussian-HS.git --recursive
• Create a conda or python environment and activate. For e.g., conda create -n anchor_gaussian python=3.9.18; conda activate anchor_gaussian.
• Install PyTorch 1.11.0 with conda or pip (instructions).
• Install PyTorch3d, we tested with 0.6.2, but any version should be fine
• Install other requirements: pip install -r requirement.txt
• Install Gaussian Splatting dependencies pip install submodules/diff-gaussian-rasterization submodules/simple-knn
• Download FLAME model, choose FLAME 2020 and unzip it, copy 'generic_model.pkl' into ./code/flame/FLAME2020

Our data format is the same as IMavatar.

Please download the subject 3 from PointAvatar at https://dataset.ait.ethz.ch/downloads/IMavatar_data/data/subject3.zip, rename subject3 to 001 and then merge with 001.zip in our released, which contains the additional DWposes needed to run our method.

To train and evaluate both the MLP and distilled version:

cd code; bash train.sh

To run reenactment:

cd code; bash reenact.sh

• Release training code
• Release reenactment code
• Release data preprocess code