Keras implementation of SRCNN

The original paper is Learning a Deep Convolutional Network for Image Super-Resolution

My implementation have some difference with the original paper, include:

• use 'he_normal' for weight initialization
• use Adam alghorithm for optimization
• I use the opencv library to produce the training data
• I did not set different learning rate in different layer, but I found this network still work.

Use:

Create your own data

open prepare_data.py and change the data path to your data

Excute: python prepare_data.py

training and test:

Excute: python main.py

Result(training for 30 epoches, with upscaling factor 2):

Method:	Bicubic	SRCNN
PSNR:	24.6971375057	28.6588428245

Origin Image:

Bicubic:

SRCNN: