DeepFillv2_Pytorch

This is a Pytorch re-implementation for the paper Free-Form Image Inpainting with Gated Convolution.

This repository contains "Gated Convolution", "Contextual Attention" and "Spectral Normalization".

Requirement

• Python 3
• OpenCV-Python
• Numpy
• Pytorch 1.0+

Compared Results

The following images are Original, Masked_orig, Official(Tensorflow), MMEditing(Pytorch), Ours(Pytorch).

Dataset

Training Dataset

The training dataset is a collection of images from Places365-Standard which spatial sizes are larger than 512 * 512. (It will be more free to crop image with larger resolution during training)

Testing Dataset

Create the folders test_data and test_data_mask. Note that test_data and test_data_mask contain the image and its corresponding mask respectively.

Training

• To train a model:

$ bash ./run_train.sh

All training models and sample images will be saved in ./models/ and ./samples/ respectively.

Testing

Download the pretrained model here and put it in ./pretrained_model/.

• To test a model:

$ bash ./run_test.sh

Acknowledgments

The main code is based upon deepfillv2.
The code of "Contextual Attention" is based upon generative-inpainting-pytorch.
Thanks for their excellent works!
And Thanks for Kuaishou Technology Co., Ltd providing the hardware support to this project.

Citation

@article{yu2018generative,
  title={Generative Image Inpainting with Contextual Attention},
  author={Yu, Jiahui and Lin, Zhe and Yang, Jimei and Shen, Xiaohui and Lu, Xin and Huang, Thomas S},
  journal={arXiv preprint arXiv:1801.07892},
  year={2018}
}

@article{yu2018free,
  title={Free-Form Image Inpainting with Gated Convolution},
  author={Yu, Jiahui and Lin, Zhe and Yang, Jimei and Shen, Xiaohui and Lu, Xin and Huang, Thomas S},
  journal={arXiv preprint arXiv:1806.03589},
  year={2018}
}