keras-examples

Image vision examples using machine learning, including

inpainting
superresolution
deblurring

All examples delivered as jupyter notebook and running on vanilla keras (with both plaidml/theano and tensorflow/google colab as backend).

results

(1. degraded image, 2. neural net output, 3. ground truth)

inpainting

superresolution

deblurring

to run...

...just upload to google/colab OR locally call git clone https://github.com/BenjaminWegener/Keras-examples

references & notes

channel attention mechanism [paper]
residual in residual architecture [paper]
subpixel convolution / pixelshuffle [paper]
running on tensorflow/google colab AND on plaidml
using the famous Set14 dataset ONLY (with heavy augmentation) - no validation needed