Usage | Web App | | Paper | Supplementary Material | More results |

code for paper ``CR-Fill: Generative Image Inpainting with Auxiliary Contextual Reconstruction". This repo (including code and models) are for research purposes only.

0. Download code

git clone --single-branch https://github.com/zengxianyu/crfill
git submodule init
git submodule update

0. Download data and model

chmod +x download/*
./download/download_model.sh
./download/download_data.sh

1. Install dependencies:

conda env create -f environment.yml

or install these packages manually in a Python 3.6 enviroment:

pytorch=1.3.1, opencv=3.4.2, tqdm, torchvision, dill, matplotlib, opencv

./test.sh

These script will run the inpainting model on the samples I provided. Modify the options --image_dir, --mask_dir, --output_dir in test.sh to test on custom data.

1. Prepare training datasets and put them in ./datasets/ following the example ./datasets/places

2. run the training script:

./train.sh

open the html files in ./output to visualize training

After the training is finished, the model files can be found in ./checkpoints/debugarr0

you may modify the training script to use different settings, e.g., batch size, hyperparameters

For finetune on custom dataset based on my pretrained models, use the following command:

1. download checkpoints

./download/download_pretrain.sh

2. run the training script

./finetune.sh

you may change the options in finetune.sh to use different hyperparameters or your own dataset

To use the web app, these additional packages are required:

flask, requests, pillow

./demo.sh

then open http://localhost:2334 in the browser to use the web app

@inproceedings{zeng2021generative,
  title={CR-Fill: Generative Image Inpainting with Auxiliary Contextual Reconstruction},
  author={Zeng, Yu and Lin, Zhe and Lu, Huchuan and Patel, Vishal M.},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  year={2021}
}

• DeepFill https://github.com/jiahuiyu/generative_inpainting
• Pix2PixHD https://github.com/NVIDIA/pix2pixHD
• SPADE https://github.com/NVlabs/SPADE