Skip to content
main
Switch branches/tags
Go to file
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 
 
 

README.md

R-MNET-A-Perceptual-Adversarial-Network-for-Image-Inpainting in Keras

R-MNET: A Perceptual Adversarial Network for Image Inpainting. Jireh Jam, Connah Kendrick, Vincent Drouard, Kevin Walker, Gee-Sern Hsu, Moi Hoon Yap

Keras implementation of R-MNET model proposed at WACV2021.

https://arxiv.org/pdf/2008.04621.pdf

Architecture

Requirements

Images dataset

Download Places2 Dataset and CelebA-HQ Dataset

Mask dataset

The training mask dataset used for training our model: QD-IMD: Quick Draw Irregular Mask Dataset
The NVIDIA's mask dataset is available here

Folder structure

After downloading the datasets, you should put create these folders into /images/train/train_images and /masks/train/train_masks. Place the images and masks in the train_images and train_masks respectively and it should be like

-- images
---- train
------ train_images
---- celebA_HQ_test
-- masks
---- train
------ train_masks
---- test_masks

/images/train/train_images and /masks/train/train_masks and place the images and masks in the train_images and train_masks respectively. Make sure the directory path is

--self.train_mask_dir='./masks/train/' 
--self.train_img_dir = './images/train/'
--test_img_dir ='./images/celebA_HQ_test/'
--test_mask_dir ='./masks/test_masks/'

Python requirements

  • Python 3.6
  • Tensorflow 1.13.1
  • keras 2.3.1
  • opencv
  • Numpy

Training and Testing scripts.

Use the run.py file to train the model and test.py to test the model. We recommend training for 100 epochs as a benchmark based on the state-of-the-art used to compare with out model.

Code Reference

  1. Wasserstain GAN was implemented based on: Wasserstein GAN Keras
  2. Generative Multi-column Convolutional Neural Networks inpainting model in Keras : Image Inpainting via Generative Multi-column Convolutional Neural Networks
  3. Nvidia Mask Dataset, based on the paper: Image Inpainting for Irregular Holes Using Partial Convolutions

Citing this script

If you use this script, please consider citing R-MNet:

@inproceedings{jam2021r,
  title={R-mnet: A perceptual adversarial network for image inpainting},
  author={Jam, Jireh and Kendrick, Connah and Drouard, Vincent and Walker, Kevin and Hsu, Gee-Sern and Yap, Moi Hoon},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={2714--2723},
  year={2021}
}
@article{jam2020r,
  title={R-MNet: A Perceptual Adversarial Network for Image Inpainting},
  author={Jam, Jireh and Kendrick, Connah and Drouard, Vincent and Walker, Kevin and Hsu, Gee-Sern and Yap, Moi Hoon},
  journal={arXiv preprint arXiv:2008.04621},
  year={2020}
}

About

R-MNET: A Perceptual Adversarial Network for Image Inpainting model in Keras

Resources

Releases

No releases published

Packages

No packages published

Languages