dreamegg / GL_Image_Inapinting_pytorch

Implementation of "Globally and Locally Consistent Image Completion"

Globally and Locally Consistent Image Completion in pytorch

  "Globally and Locally Consistent Image Completion"
  Satoshi Iizuka, Edgar Simo-Serra, and Hiroshi Ishikawa
  ACM Transaction on Graphics (Proc. of SIGGRAPH 2017), 2017

This is an implementation of the image completion model proposed in the paper (Globally and Locally Consistent Image Completion) in Pytorch 0.4.

Requirements

• Python 3
• Pytorch 0.4
• TensorbardX
• argparser
• etc (PIL, tqdm...)

Usage

I. Prepare the training data

This step is pre-pocessing of the image (make random mask) and transform image to torch tensor.

$ cd src_gl
$ python prepare_dataset2tensor.py

in github. I already uploaded "Facade dataset" with test and train. You can put any of datasets in data folder

And also you can change image size, maske size, maske margine

$ python prepare_dataset2tensor.py --patchSize 256 --patchSize 128 --boundary_margin 8

II. Train model

Train the "GL" model with pre-processed tensor data in step I.

$ cd src_gl
$ python train.py

III. Evaluate model

$ cd src_gl
$ python eval.py

Result

Related works

• Author Code (https://github.com/satoshiiizuka/siggraph2017_inpainting)
• GLCIC in tnesorflow (https://github.com/tadax/glcic)
• Code Base (https://github.com/pytorch/examples/tree/master/super_resolution)

Citation

Please cite the original paper in your publications:

@Article{IizukaSIGGRAPH2017,
  author = {Satoshi Iizuka and Edgar Simo-Serra and Hiroshi Ishikawa},
  title = {{Globally and Locally Consistent Image Completion}},
  journal = "ACM Transactions on Graphics (Proc. of SIGGRAPH)",
  year = 2017,
  volume = 36,
  number = 4,
  pages = 107:1--107:14,
  articleno = 107,
}

#Implementation Author chankyoo.moon / @dreamegg