GrayScale-SRCNN

Data Preprocessing

Use bilinear interpolation to transfer the 64x64 images to 128x128 images for both train and test

python3 preprocess.py

Train

Train by SRCNN

Use SRCNN model to train the 16000 images with 500 epoch, 1 batch and 1e-4 learning rate (about 10 hours)

python3 train_srcnn.py

Train by VDSR

Use VDSR model to train the 16000 images with 500 epoch, 1 batch and 1e-4 learning rate (about 50 hours)

python3 train_vdsr.py

Test

Load the final_train_model.pth (trained by VDSR) to output 3999 test images

python3 output.py