Iterative Facial Image Inpainting

This repository contains source codes and training sets for the following paper:

"Iterative Facial Image Inpainting Based on an Encoder-Generator Architecture" by Yahya DOGAN and Hacer YALIM KELES.

The preprint version of the above paper is available at: https://arxiv.org/abs/2101.07036

Requirements

• Spyder 4.2.x (recommended)
• Python 3.6
• Keras 2.2.4
• Tensorflow-gpu 1.13.1

Usage

1. Clone this repo: git clone https://github.com/yahyadogan72/iterative_facial_image_inpainting.git
2. Download CelebA dataset (note: convert delta files to png format), then put it in the following directory structure:
   |-Iterative_facial_image_inpainting
   |---Discriminator
   |----dataset/real
3. Prepare the dataset for the discriminator network using the script.
4. Train the discriminator network using the script.
5. Download the pre-trained models and put them in the pretrained _models folder.
6. Create dataset for the Unet model using the CRG model, then put it in the following directory structure:
   |-Iterative_facial_image_inpainting
   |--Unet
   |---dataset
   |----train/input
7. Train the Unet network using the script.
8. Test the overall model using the script.
9. You can use the script to create a mask that includes all facial landmarks in a face or download test data here.

Citation

If you find iterative facial image inpainting method useful in your research, please consider citing:

Dogan, Y. & Keles, H.Y. Neural Comput & Applic (2022). https://doi.org/10.1007/s00521-022-06987-y

Preprint:

@article{dogan2021iterative,
  title={Iterative Facial Image Inpainting using Cyclic Reverse Generator},
  author={Dogan, Yahya and Keles, Hacer Yalim},
  journal={arXiv preprint arXiv:2101.07036},
  year={2021}
}

Contact

yahyadogan72 at gmail.com