SRCNN

Download weights, and extract to the default ./Results/

python prepare_data.py --filter srcnn
# Download test data (optional)
python prepare_data.py --filter set5 --data_dir=./Data

Test model:

cd Train
# use PIL's bicubic upsample to LR, by `-f=scale#4`, where `#4` passes parameter `4` to callback function `scale(int)`
python run.py --model=srcnn --test=set5 --custom_upsample=true -f=scale#4

Expected output

INFO:tensorflow:Restoring parameters from ../Results/srcnn/save/srcnn-sc4-ep0200.ckpt

Test: 100%|##################| 5/5 [00:02<00:00, 2.21it/s]

mse: 128.624557, psnr: 28.477646, ssim: 0.826166 \n