This is a repository for Single Image (Deep) Super Resolution approaches.
Models Included: SRCNN, Baisic/Advanced DRCNN
Model_Scripts Folder:
- models.py - file containing all model classes and custom loss functions
- utils.py - all utility functions for things like creating tile dataset, stiching images back together, etc -(SRCNN_Lab.py, DRC_Basic.py, DRC_Advanced.py) - Model Training/Testing Loops
Image_Scripts Folder:
- Generate_LR_Tiled_Fixed.py - Creates Tiled Dataset for Early Upsample Training
- make_gif.py - used for creating demo gifs
Currently the best results are for SRCNN with a Skip connection. Below are a comparison of SRCNN at 400 Epochs on Adam vs SRCNN w/ Skips at 180 Epochs on Adam
Gif of Test Results Every 10 Epochs
SRCNN W/ Skip Metrics
SRCNN W/ Skip Re-Tiled Outputs
SRCNN W/ Skip Re Tiled Side by Side
