FFaceNeRF

[CVPR2025] FFaceNeRF: Few-shot Face Editing in Neural Radiance Fields

[Project Page] [Paper]

⚙️ Install Environment via Anaconda (Recommended)

conda env create -f environment.yml
conda activate ffacenerf

FFaceNeRF requires NeRRFaceEditing checkpoints for intialization

put pretrained_model into networks/NeRFFaceEditing-ffhq-64.pkl

Download Data for training and testing

Training requires about 40 minutes on single A6000 GPU

python train_ffacenerf.py --mode eyes
#python train_ffacenerf.py --mode nose
#python train_ffacenerf.py --mode chin

Testing

python editing_testset.py --mode eyes --network ckpt_eyes_10.pth --overlap_weight 0.5
#python editing_testset.py --mode eyes --network ckpt_nose_10.pth --overlap_weight 0.4 --target_image 64

In the original CVPR paper, the test set comprised 22 samples; in this public repository, we expanded it to 41 for the testing.

python evaluate.py

We would like to thank EG3D and NeRFFaceEditing for open-source video interpolation model

Citation

@inproceedings{yun2025ffacenerf,
  title={FFaceNeRF: Few-shot Face Editing in Neural Radiance Fields},
  author={Yun, Kwan and Kim, Chaelin and Shin, Hangyeul and Noh, Junyong},
  booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference},
  pages={10825--10835},
  year={2025}
}