Torch Implementation of "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network" [Paper]
This is a prototype implementation developed by Harry Yang.
####Training prepare your images under a sub-folder of a root folder
train_folder=your_root_folder model_folder=your_save_folder th run_sr.lua
<<<<<<< HEAD
By default it resizes the images to 96x96 as ground truth and 24x24 as input, but you can specify the size using loadSize
. Note current generator network only supports 4x super-resolution.
By default it resizes the images to 96x96 as ground truth and 24x24 as input, but you can specify the size using loadSize
and scale
.
2e06dcd73670f06c9160101f0eb9753a6e81d841
####Loading a saved model to train
D_path=your_saved_D_model G_path=your_saved_G_model train_folder=your_root_folder model_folder=your_save_folder th run_resume.lua
####Testing prepare your test images under a sub-folder of a root folder
test_folder=your_root_folder model_file=your_G_model result_path=location_to_save_results th run_test.lua
##Todo:
- Add TV loss.
- Provide trained models.