This is the official source code for the paper "Realistic Full-Body Anonymization with Surface-Guided GANs".

[Arixv Paper] [Appendix] [Google Colab demo] [WACV 2023 Conference Presentation]

Surface-guided GANs is an automatic full-body anonymization technique based on Generative Adversarial Networks.

The key idea of surface-guided GANs is to guide the generative model with dense pixel-to-surface information (based on continuous surface embeddings). This yields highly realistic anonymization result and allows for diverse anonymization.

• Pytorch >= 1.9
• Torchvision >= 0.11
• Python >= 3.8
• CUDA capable device for training. Training was done with 1-4 32GB V100 GPUs.

We recommend to setup and install pytorch with anaconda following the pytorch installation instructions.

1. Clone repository: git clone https://github.com/hukkelas/full_body_anonymization/.
2. Install using setup.py:

pip install -e .

Otherwise, you can setup your environment with our provided Dockerfile.

The file anonymize.py can anonymize image paths, directories and videos. python anonymize.py --help prints the different options.

To anonymize, visualize and save an output image, you can write:

python3 anonymize.py configs/surface_guided/configE.py coco_val2017_000000001000.jpg --visualize --save

The truncation value decides the "creativity" of the generator, which you can specify in the range (0, 1). Setting -t 1 will generate diverse anonymization between runs. For config A/B/C, the truncation value accepts range of (0, $\infty$). Setting -t=None will apply to latent truncation.

python3 anonymize.py configs/surface_guided/configE.py coco_val2017_000000001000.jpg --visualize --save -t 1

Check out the interactive demo with our gradio implementation. Run

python3 app.py

See docs/TRAINING.md.

See docs/REPRODUCING.md.

All code, except the stated below, is released under MIT License.

Code under:

• torch_utils/: Code modified from github.com/NVlabs/stylegan2-ada-pytorch. Separate license is attached in the directory.
• dnnlib/: Code modified from github.com/NVlabs/stylegan2-ada-pytorch. Separate license is attached in the directory.
• Detection network: See Detectron2 License.

If you use this code for your research, please cite:

@inproceedings{hukkelas23FBA,
  author={Hukkelås, Håkon and Smebye, Morten and Mester, Rudolf and Lindseth, Frank},
  booktitle={2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)}, 
  title={Realistic Full-Body Anonymization with Surface-Guided GANs}, 
  year={2023},
  volume={},
  number={},
  pages={1430-1440},
  doi={10.1109/WACV56688.2023.00148}}