Localization of Deep Inpainting Using High-Pass Fully Convolutional Network

This is the implementation of the paper Localization of Deep Inpainting Using High-Pass Fully Convolutional Network (ICCV 2019).

Requirements

• Python 3
• Tensorflow >= 1.10.0

Usage

Train

First, prepare the training data so that the images are stored in "xxx/jpg/xxx/" and the corresponding groundtruth masks are stored in "xxx/msk/xxx/". Then, run the following command.

python3 hp_fcn.py --data_dir <path_to_the_training_dataset> --logdir <path_to_the_log_directory_for_saving_model_and_log> --mode train

Test

Prepare the testing data in a similar way and run the code as follows.

python3 hp_fcn.py --data_dir <path_to_the_testing_dataset> --logdir <path_to_the_log_directory> --restore <path_to_the_trained_model> --mode test

Pretrained checkpoint

The pretrained checkpoint of High-pass FCN is available at: https://drive.google.com/drive/folders/1W1f_piFIiK6JJLIXimr1vtRs8MVYLwjZ?usp=sharing

Dataset

The training and testing images inpainted with the method "Globally and locally consistent image completion" (Iizuka et al., TOG 2017) are available at: https://pan.baidu.com/s/1Df77EOoBkhukLNAz3YKT4w?pwd=gi60

Note

This repo also includes an implementation of MFCN (mfcn.py):

Ronald Salloum, Yuzhuo Ren, and C.-C. Jay Kuo. Image splicing localization using a multi-task fully convolutional network (MFCN). Journal of Visual Communication and Image Representation, 51:201–209, 2018.

To run the code of MFCN, you should also have the edge masks in "xxx/edg*/xxx/".

Citation

If you use our code please cite:

@InProceedings{Li_2019_ICCV,
    author = {Li, Haodong and Huang, Jiwu},
    title = {Localization of Deep Inpainting Using High-Pass Fully Convolutional Network},
    booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
    pages={8301--8310},
    month = {October},
    year = {2019}
}

Help

If you have any questions, please contact: lihaodong[AT]szu.edu.cn