CS766_Project

Webpage

Here to find out more about the project.

Requirements

pytorch >= 1.8.0 with cuda pytorch_msssim pandas scikit-image

How to use

1. Prepare the SR data set From SRDataset.py documentation: An SRDataset is a directory that includes the following components:

        /files.csv -- contains a list of file names, without directory path
        /img/ -- the directory where the original images are stored
        /hr/ -- the HR images, used as input, they are scaled and cropped from the original images
   
        /lr2x/, /lr4x/, /lr8x/ -- the LR images
        /edge/ -- the edge generated from HR images
        /edge_lr2x/,/edge_lr4x/,/edge_lr8x/ LR edge from LR images
   
        /pred_edge_lr2x/, etc. -- predicted edges
        /pred_full_lr2x/, etc. -- the SR images
   
        Note: Some images should be generated before use. For example, to train only the SR model, the predicted edge images(pred_edge_lr2x, etc.) should be generated beforehand.
    Each line in files.csv correspond to a file in `img/`
   

2. Specify the specs: the path to the data, the path of the trained model etc.

   • Refering to config.json for more information.

3. Run python3 main.py [config file name]

   • By default config.json is used