Paper | Project page | Demo video

Official implementation of "Make Encoder Great Again in 3D GAN Inversion through Geometry and Occlusion-Aware Encoding" ICCV 2023

We present a encoder-based 3D generative adversarial network (GAN) inversion framework that can efficiently synthesize photo-realistic novel views while preserving geometry and details of the input image.

• 64-bit Python 3.8 and PyTorch 1.11.0 (or later).
• CUDA toolkit 11.3 or later.
• Python libraries: see requirements.txt

cd goae
conda create --name goae python=3.8
conda activate goae
pip install -r requirements.txt

Dataset preparation and new test sample camera processing can refer to EG3D or SPI preprocess

The pretrained model checkpoint can be downloaded from google drive, Put those checkpoint into the directory GOAE/pretrained . Note that current pretrained AFA only modifies the triplane on 32*32 resolution, more higher resolution modify can achieve better result.

You can use the command below to test the example.

python infer.py --multi_view --video

You can use the command below to edit the example.

python infer.py --multi_view --video --edit --edit_attr glass --alpha 1.0

Training codes can be downloaded from here.

If you find this work useful for your research, please cite:

@article{yuan2023make,
  title={Make Encoder Great Again in 3D GAN Inversion through Geometry and Occlusion-Aware Encoding},
  author={Yuan, Ziyang and Zhu, Yiming and Li, Yu and Liu, Hongyu and Yuan, Chun},
  journal={arXiv preprint arXiv:2303.12326},
  year={2023}
}

If you have any comments or questions, please open a new issue or feel free to contact Ziyang Yuan (yuanzy22@mails.tsinghua.edu).