Super resolution based on SRCNN using Keras (2.0)
Please refer to paper for more details on working https://arxiv.org/pdf/1501.00092v3.pdf
Install require dependency
pip install Pillow==2.2.2
pip install tensorflow
Image dataset was downloaded from google images, refer to link
(https://www.pyimagesearch.com/2017/12/04/how-to-create-a-deep-learning-dataset-using-google-images/)
Two architectures are implemented.
1. Expanded Super Resolution CNN (ESRCNN)
This gave good results,
2. Denoiseing Super Resolution CNN (DSRCNN)
This had better results than previous.
Both the model was trained on 2000 images for 500 epochs. In this notebook I have used 128x128 size images
Credits
Architecture details - https://github.com/olgaliak/