Official pytorch impementation of our private-encoder paper for both training and inferencing.
- Install required packages
wget https://github.com/ninja-build/ninja/releases/download/v1.8.2/ninja-linux.zip
sudo unzip ninja-linux.zip -d /usr/local/bin/
sudo update-alternatives --install /usr/bin/ninja ninja /usr/local/bin/ninja 1 --force
- Download required pretrained models
The auxiliary models which are needed for training your own model from scratch or inference.
| Path | Description |
|---|---|
| FFHQ StyleGAN | StyleGAN model pretrained on FFHQ taken from rosinality with 1024x1024 output resolution. |
| IR-SE50 Model | Pretrained Arcface for use in our ID loss during training. |
| MTCNN | Weights for MTCNN model taken from TreB1eN for use in ID similarity metric computation. (Unpack the tar.gz to extract the 3 model weights.) |
Please download and save these models to the directory pretrained_models.
- The main training script can be found in
scripts/train.py. - Before training, please define your data path at
--train_source_rootand--test_source_root. - The training results are saved to
exp_dir=./experiment, so you can check the checkpoints, tensorboard logs, intermediate train output and test outputs in that directory. - To acquire different anonymization effects, you can adjust
--target_id.
python scripts/train.py \
--exp_dir=./experiment \
--dataset_type=celeba_encode \
--checkpoint_path=./pretrained_models/psp_ffhq_encode.pt \
--train_source_root=/content/drive/MyDrive/Image_datasets/celeba_align_20k_21k_aligned \
--test_source_root=/content/drive/MyDrive/Image_datasets/celeba_align_20k_21k_aligned \
--workers=8 \
--batch_size=4 \
--test_workers=8 \
--test_batch_size=4 \
--encoder_type=GradualStyleEncoder \
--start_from_latent_avg \
--lpips_lambda=0.8 \
--id_lambda=0.1 \
--l2_lambda=1.0 \
--max_steps=100000 \
--image_interval=5000 \
--board_interval=50 \
--val_interval=1000 \
--save_interval=5000 \
--target_id=0.0
You can also directly run training colab notebook example in
- Pretrained Models
Please download the pre-trained Private-Encoder models from the following links.
| Path | Description |
|---|---|
| Pretrained Models for FFHQ | trained with the FFHQ dataset for different level of identity anonymization. |
- If you wish to use one of the pretrained models for training or inference, please change the directory of
--checkpoint_path. - Before inference, please define your data path at
--data_path.
python scripts/inference.py \
--exp_dir=./image_datasets \
--checkpoint_path=./best_models/privacy_0.0/best_model.pt \
--data_path=./image_datasets/ffhq5k_6k_related/ffhq5k_6k_aligned \
--test_batch_size=5 \
--test_workers=4
You can also directly run inference colab notebook example in
If you use this code for your research, please cite our paper Private-Encoder: Enforcing Privacy in Latent Space for Human Face Images:
@article{zhao2021private,
title={Private-encoder: Enforcing privacy in latent space for human face images},
author={Zhao, Yuan and Liu, Bo and Zhu, Tianqing and Ding, Ming and Zhou, Wanlei},
journal={Concurrency and Computation: Practice and Experience},
pages={e6548},
year={2021},
publisher={Wiley Online Library}
}