PA-GAN

PA-GAN: Progressive Attention Generative Adversarial Network for Facial Attribute Editing
Zhenliang He^1,2, Meina Kan^1,2, Jichao Zhang^1,2,3, Shiguang Shan^1,2
1Key Lab of Intelligent Information Processing, Institute of Computing Technology, CAS, China
²University of Chinese Academy of Sciences, China
³Peng Cheng Laboratory, China

Usage

• Environment

  • Python 3.6

  • TensorFlow 1.13+, TensorFlow Graphics

  • OpenCV, scikit-image, tqdm, oyaml

  • we recommend Anaconda or Miniconda, then you can create the PA-GAN environment with commands below

    conda create -n PA-GAN python=3.6
    
    source activate PA-GAN
    
    conda install opencv scikit-image tqdm tensorflow-gpu=1.13
    
    conda install -c conda-forge oyaml
    
    pip install tensorflow-graphics-gpu --no-deps

  • NOTICE: if you create a new conda environment, remember to activate it before any other command

    source activate PA-GAN

• Data Preparation

  • CelebA-unaligned (10.2GB, higher quality than the aligned data)

    • download the dataset

      • img_celeba.7z (move to ./data/img_celeba/img_celeba.7z): Google Drive or Baidu Netdisk (password rp0s)

      • annotations.zip (move to ./data/img_celeba/annotations.zip): Google Drive

    • unzip and process the data

      7z x ./data/img_celeba/img_celeba.7z/img_celeba.7z.001 -o./data/img_celeba/
      
      unzip ./data/img_celeba/annotations.zip -d ./data/img_celeba/
      
      python ./scripts/align.py

• Run PA-GAN

  • training

    CUDA_VISIBLE_DEVICES=0 \
    python train.py \
    --experiment_name PA-GAN_128

  • testing

    • single attribute editing (inversion)

      CUDA_VISIBLE_DEVICES=0 \
      python test.py \
      --experiment_name PA-GAN_128

    • multiple attribute editing (inversion) example

      CUDA_VISIBLE_DEVICES=0 \
      python test_multi.py \
      --test_att_names Bushy_Eyebrows Mustache \
      --experiment_name PA-GAN_128

  • loss visualization

    CUDA_VISIBLE_DEVICES='' \
    tensorboard \
    --logdir ./output/default/summaries \
    --port 6006

• Using Trained Weights

  • alternative trained weights (move to ./output/*.zip)

    • PA-GAN_128.zip (953.9MB)

      • including G, D, and the state of the optimizer

    • PA-GAN_128_generator_only.zip (151.5MB)

      • G only

    • PA-GAN_256_generator_only.zip (70.8MB)

  • unzip the file (PA-GAN_128.zip for example)

    unzip ./output/PA-GAN_128.zip -d ./output/

  • testing (see above)

Citation

If you find PA-GAN useful in your research work, please consider citing:

@article{he2020pagan,
  title={PA-GAN: Progressive Attention Generative Adversarial Network for Facial Attribute Editing},
  author={He, Zhenliang and Kan, Meina and Zhang, Jichao and Shan, Shiguang},
  journal={arXiv preprint arXiv:2007.05892},
  year={2020}
}