Keras implementation of EEDS: End-to-End image super-resolution via deep and shallow convolutional networks
The original paper is end-to-end image super-resolution via deep and shallow convolutional networks
This implementation only focuse on two-scales' upsampling, and it is easy to change this code to perform upscaling with other ratio by change the strides in Conv2DTranspose. And because of the difference between the color space of YCrCb in Matlab and OpenCV, the final PSNR will have some difference. Results evluated with the MATLAB code will be a little higher.
Excute:
python EEDS_predict_demo.py
open prepare_data.py and change the data path to your data
Excute:
python prepare_data.py
Excute:
python EEDS.py
Excute:
python EES.py
EEDS and EES training for 200 epoches, with upscaling factor 2
Origin Image:
input:
EES:
EEDS:
After conv1:
After Deconv:
This code of feature map visualization part refered the code of https://www.kaggle.com/abhmul/leaf-classification/keras-convnet-lb-0-0052-w-visualization