Deep autoencoder with skip connection neural networks, which also called U-Net.
This project is based on Denoise-VAE
The constructure:
U-Net.ipynb uses keras with tensorflow backend.
The six pictures in first row are the original pictures.
The second are the damaged pictures which rows of 10 to 18 were coverd by zeros.
The third are the results of the U-Net.ipynb
git clone https://github.com/block98k/U-Net-keras.git
cd ./U-Net-keras
jupyter notebook U-Net.ipynb
tensorflow 1.0 or above
numpy
jupyter
This is a project I used to practice autoencoder and U-Net.