Semantic-Guided Face Inpainting with Subspace Pyramid Aggregation

prepare datasets

• Download public image datasets or your own image datasets and download the public masks.
• Split your dataset into training set, test set and validation set according to the ratio of 8:1:1.
• Specify the path to training data by --dir_image and --dir_mask.

Requirements

• python 3.8.8
• pytorch (tested on Release 1.8.1)

Installation

• git clone https://github.com/xiumin123/Face_inpainting.git
• cd Face_inpainting
• conda env create -f environment.yml
• conda activate inpainting

Getting Started

• download pretrained model,place this model under the floder src/model/model_pr

  https://pan.baidu.com/s/1pHgGfYht8vKq1tJyEpE6Iw 提取码: 7r9i

• train:

  • Before you train the model, you can choose the model variants in the "src/model" folder. You can modify the necessary parameters in "src/utils/options.py".
  • cd src
  • python train.py

• test:

  • cd src
  • python test.py --pre_train [path to pretrained model]

• Evaluating:

  • cd src
  • python eval.py --real_dir [ground truths] --fake_dir [inpainting results] --metric mae psnr ssim fid

Acknowledgement

Our models were trained and tested on an RXT3090Ti GPU and Intel(R) Xeon(R) Silver 4214 CPU @ 2.20GHz.All the experimental data in this thesis were reproduced by myself on the server.Further more, our experiment is developed relying on AOT-GAN and other projects. Thanks for these great projects.