Keras implementation of Colorful Image Colorization by Richard Zhang, Phillip Isola and Alexei A. Efros
The technique is applied on the CelebA dataset with minor modifications.
- scikit-learn==0.19.1
- scikit-image==0.13.1
- tqdm==4.17.0
- opencv_python==3.3.0.10
- numpy==1.13.3
- matplotlib==2.0.2
- Keras==2.0.8
- Theano==0.9.0 or Tensorflow==1.3.0
- h5py==2.7.0
- parmap==1.5.1
- scipy==1.0.0
- Nvidia GPU with at least 2GB RAM
- At least 4GB RAM (when using the on_demand option for training)
The settings above should work well enough for small image size (32 x 32). Above that, better GPU and more RAM are required.
Follow these instructions.
Follow these instructions
For each triplet:
- First column is the original
- Second column is the B&W version
- Last column is the colorized output
Follow these instructions