http://personal.ie.cuhk.edu.hk/~ccloy/files/eccv_2014_deepresolution.pdf
This is an implementation of SRCNN using keras. You can conduct simple experiments of super resolution on Set5 or your own images.
- python 3.6
- keras
- opencv
python train.py
To conduct using GPU is recommended. There is a pretrained model in './model', so you don't have to train model.
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