srcnn_keras

http://personal.ie.cuhk.edu.hk/~ccloy/files/eccv_2014_deepresolution.pdf

Abstract

This is an implementation of SRCNN using keras. You can conduct simple experiments of super resolution on Set5 or your own images.

Requirements

• python 3.6
• keras
• opencv

Usage

Train

python train.py
To conduct using GPU is recommended. There is a pretrained model in './model', so you don't have to train model.

Test

python test.py
Outputs are restored in './result'. If you want to try super resolution to your own images, please put them in './test'.

• original: images preprocessed as downsampling 1/scale (default scale=3)
• answer: target images (raw images in './test')
• input: 'original' images preprocessed as upsampling scale/1 (default scale=3)
• predicted: outputs images

References

kweisamx/TensorFlow-SRCNN