Our approach to the SR problem by building upon the work of the following papers and by providing our model with extra loss functions and a diverse set of weight parameters.
- Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network
- A U-Net Based Discriminator for Generative Adversarial Networks
- Perceptual Losses for Real-Time Style Transfer and Super-Resolution
You can download DIV2K dataset, from the following links : train_HR, train_LR_bicubic_X4, valid_HR, valid_LR_bicubic_X4.
Your directory structure for the dataset should look like this:
DIV2K
└── DIV2K_train_HR
├── DIV2K_train_LR_bicubic
├── DIV2K_valid_HR
└── DIV2K_valid_LR_bicubic
