Official pytorch implementation for "GSHeadRelight: Fast Relightability for 3D Gaussian Head Synthesis".
For jittor implementation please see (https://github.com/IGLICT/GSHeadRelight_jittor).
- (25/08/14) Release inference/training code and pretrained checkpoint (https://drive.google.com/file/d/1ZCLThyUj73Je-Set6yHc7L_-IbindzJm).
We build our code on GSGAN (https://github.com/hse1032/GSGAN). Run below code snippet for installing dependencies.
conda env create -f environment.yml
conda activate gsheadrelight
git clone --recursive https://github.com/ashawkey/diff-gaussian-rasterization
pip install ./diff-gaussian-rasterization
imageio_download_bin freeimage
Pretrained checkpoint is available in Google Drive link (https://drive.google.com/file/d/1ZCLThyUj73Je-Set6yHc7L_-IbindzJm).\
Please download it and put it in models/network-snapshot.pkl.
We exactly follow the data-preprocessing of EG3D, so please refer to them (https://github.com/NVlabs/eg3d).
We provide the training script in "scripts/train{dataset}.sh" file. Only need to change is the "dataset_path" variable. Training results are saved at "outdir" directory.
sh _script/train_ffhq512.sh # training on FFHQ512
We provide the evaluation of FID.
Set the path of "network_pkl" in "_scripts/eval.sh" file properly, and run it.
Or you can simply run below code for evaluation.
python calc_metrics.py --metrics=fid50k_full --network=${network_pkl}
We provide three scripts for generating videos with different settings.
sh _script/generate_videos_sh.sh # generate videos with environment map
sh _script/generate_videos_envmap.sh # generate videos with environment map
sh _script/generate_videos_albedo.sh # generate videos of albedo
Set the path of "network_pkl" in "scripts/generate_videos*.sh" file properly.
Our code is based on the following repositories.
GSGAN (https://github.com/hse1032/GSGAN)
EG3D (https://github.com/NVlabs/eg3d)
Gaussian-Splatting (https://github.com/graphdeco-inria/gaussian-splatting)
dreamgaussian (https://github.com/dreamgaussian/dreamgaussian)
For any further questions, please contact lvhenglei22s@ict.ac.cn
If our paper or codebase is useful in your research, please cite our paper!
@inproceedings{lv2025gsheadrelight,
title={GSHeadRelight: Fast relightability for 3D gaussian head synthesis},
author={Lv, Henglei and Deng, Bailin and Guo, Jianzhu and Liu, Xiaoqiang and Wan, Pengfei and Zhang, Di and Gao, Lin},
booktitle={Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers},
pages={1--12},
year={2025}
}