Region Normalization for Image Inpainting

The paper can be found here.

The codes are initial version, not the revised version. I will update these someday since I am busy currently. However, if you have any question about the paper/codes, you can contact me through Email([email protected]).

Please run the codes where the python is Version 3.x and pytorch>=0.4.

Preparation

Before running the codes, you should prepare training/evaluation image file list (flist) and mask file list (flist). You can refer to the folowing command to generate .flist file:

python flist.py --path your_dataset_folder --output xxx.flist

Training

There are some hyperparameters that you can adjust in the main.py. To train the model, you can run:

python main.py --bs 14 --gpus 2 --prefix rn --img_flist your_training_images.flist --mask_flist your_training_masks.flist

PS: You can set the "--bs" and "--gpus" to any number as you like. The above is just an example.

Evaluation

To evaluate the model, you can use GPU or CPU to run.

For GPU:

python eval.py --bs your_batch_size --model your_checkpoint_path --img_flist your_eval_images.flist --mask_flist your_eval_masks.flist

For CPU:

python eval.py --cpu --bs your_batch_size --model your_checkpoint_path --img_flist your_eval_images.flist --mask_flist your_eval_masks.flist

PS: The pretrained model under folder './pretrained_model/' is trained from Places2 dataset with Irregular Mask dataset. Please train RN from scratch if you test data not from Places2 or using regular mask.

Cite Us

@misc{yu2019region,
    title={Region Normalization for Image Inpainting},
    author={Tao Yu and Zongyu Guo and Xin Jin and Shilin Wu and Zhibo Chen and Weiping Li and Zhizheng Zhang and Sen Liu},
    year={2019},
    eprint={1911.10375},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

Appreciation

The codes refer to EdgeConnect. Thanks for the authors of it！