If you want to download this code, please use command:
git clone https://github.com/GuardSkill/Large-Scale-Feature-Inpainting.git

BibTex

• Python 3.6
• PyTorch 1.0 （The version MUST >=1.0）
• NVIDIA GPU + CUDA cuDNN

• Clone this repo:

git clone https://github.com/GuardSkill/Large-Scale-Feature-Inpainting.git
cd Large-Scale-Feature-Inpainting

• Install PyTorch.
• Install python requirements:

pip3 install -r requirements.txt

We use Places2, CelebA datasets. To train a model on the full dataset, download datasets from official websites.

After downloading, run scripts/flist.py to generate train, test and validation set file lists for images or masks. To generate the training set file lists on Places2 dataset run:

mkdir datasets
python3 ./scripts/flist.py --path [path_to_places2_train_set] --output ./datasets/places2_train.flist
python3 ./scripts/flist.py --path [path_to_places2_validation_set] --output ./datasets/places2_val.flist
python3 ./scripts/flist.py --path [path_to_places2_test_set] --output ./datasets/mask_test.flist

We alse provide the function for generate the file lists of CelebA by using the official partition file. To generate the train,val,test dataset file lists on celeba dataset run:

python3 ./scripts/flist.py --path [path_to_celeba_dataset] --celeba [path_to_celeba_partition_file] 

Our model is trained on the noised irregular mask dataset(.jpg format, 0 represents invalid region), which an be download at the Chinese NetDisk-坚果云. These jpeg masks are made from Irregular Mask Dataset provided by Liu et al..

We additionally provide the code for dividing the mask maps into to 4 class according to proportion of their corrupted region.

Download the pre-trained models using the following links and copy them under ./checkpoints directory.

Pretrained on Places2: mega | Google Drive | 坚果云

Pretrained on CelebA: mega | Google Drive

To train the model, create a config.yaml file similar to the example config file and copy it under your checkpoints directory. Read the configuration guide for more information on model configuration.

To train the ISNet:

python3 train.py --checkpoints [path to checkpoints]

To test the model, create a config.yaml file similar to the example config file and copy it under your checkpoints directory. Read the configuration guide for more information on model configuration.

You can evaluate the test dataset which list file path is recorded in config.yaml by this command:

python3 test.py \
--path ./checkpoints/Celeba

You can test the model for some specific images and masks, you need to provide an input images and a binary masks. Please make sure that the resolution of mask is same as images To test the model:

python3 test.py \
  --checkpoints [path to checkpoints] \
  --input [path to input directory or file] \
  --mask [path to masks directory or mask file] \
  --output [path to the output directory]

We provide some test examples under ./examples directory. Please download the pre-trained models and run:

python3 test.py \
  --checkpoints ./checkpoints/places2 
  --input ./examples/places2/images 
  --mask ./examples/places2/masks
  --output ./examples/places2/results

This script will inpaint all images in ./examples/places2/images using their corresponding masks in ./examples/places2/mask directory and saves the results in ./checkpoints/places2/results directory.

The model configuration is stored in a config.yaml file under your checkpoints directory. The following tables provide the documentation for all the options available in the configuration file:

Licensed under a Creative Commons Attribution-NonCommercial 4.0 International.

Except where otherwise noted, this content is published under a CC BY-NC license, which means that you can copy, remix, transform and build upon the content as long as you do not use the material for commercial purposes and give appropriate credit and provide a link to the license.

Lots of logic code and readme file comes from Edge-Connect, we sincerely thanks their contribution.

If you use this code for your research, please cite our paper Interactive Separation Network For Image Inpainting:

@inproceedings{li2020interactive,
  title={Interactive Separation Network For Image Inpainting},
  author={Li, Siyuan and Lu, Lu and Zhang, Zhiqiang and Cheng, Xin and Xu, Kepeng and Yu, Wenxin and He, Gang and Zhou, Jinjia and Yang, Zhuo},
  booktitle={2020 IEEE International Conference on Image Processing (ICIP)},
  pages={1008--1012},
  year={2020},
  organization={IEEE}
}