Train a recurrent network to generate grayscale images using Keras in Python 3.6
One set of weights included, from training over a set of 46 Japanese inkwash (sumi-e) artworks. Training dataset not included.
Easily adapt to train on your own grayscale image set and generate new images. Recommend training on a GPU.
In order to generate images using my pre-trained model, you'll only need the weights-improvement-00-0.4125-3.hdf5 and sample.py files. You'll need to supply sample.py with at least one grayscale image from which the prediction model can generate a seed. Enter the path to this image or a folder of images in line 19.
- Manuel Garrido and his keras_monet project
- Jason Brownlee's tutorial for text generation with LSTMs in Keras