Pytorch-Implement-Faster-High-Res-Neural-Inpainting

This is a Pytorch implementation for paper "High-Resolution Image Inpainting using Multi-Scale Neural Patch Synthesis"

Version

• python 3.6/3.7

• pytorch 1.1.0

• torchvision 0.3.0

• opencv-python 4.2.0.32

Examples

This is the python code for High-Resolution Image Inpainting using Multi-Scale Neural Patch Synthesis. The code is adapted from Faster-High-Res-Neural-Inpainting. Given an image, we use the content and texture network to jointly infer the missing region.

Demo

• Download the pre-trained models (trained on 6000 pictures from Paris StreetView for 25 epoches) for the content and texture networks and put them under the folder model/.

• Run the Demo

  cd Pytorch-Implement-Faster-High-Res-Neural-Inpainting
  # This will use the trained model to generate the output
  python run_your_pic.py --content_path "For_test/001101_2.jpg" (Path of your picture)
  # Because sample models we provided was trained on 6000 pictures from dataset Paris StreetView,
  # We recommend that you use pictures with street views to run the demo.
  # For your convenience, we provide Street pictures not in the training set for you to run the 
  # demo in the folder "For_test"

• The results will be in the folder "pic_result" which including some intermediate results. The final reulst will be named as "result".

Reference

[1]. Yang, Chao and Lu, Xin and Lin, Zhe and Shechtman, Eli and Wang, Oliver and Li, Hao. High-Resolution Image Inpainting using Multi-ScaleNeural PatchSynthesis[C].//The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017