First attempt to use the previous CPPN-GAN-VAE model to train on CIFAR-10 images.
Python
Latest commit f4a8db1 Apr 6, 2016 HaDavid first commit
Permalink
Failed to load latest commit information.
cifar-10-batches-py first commit Apr 6, 2016
examples first commit Apr 6, 2016
save first commit Apr 6, 2016
README.md first commit Apr 6, 2016
cifar.py first commit Apr 6, 2016
images2gif.py first commit Apr 6, 2016
model.py first commit Apr 6, 2016
ops.py first commit Apr 6, 2016
sampler.py first commit Apr 6, 2016
train.py first commit Apr 6, 2016

README.md

cppn-gan-vae tensorflow on CIFAR-10

Train Compositional Pattern Producing Network as a Generative Model, using Generative Adversarial Networks and Variational Autoencoder techniques to produce high resolution images.

CIFAR Truck CPPN Output after training on the Truck class of CIFAR-10

sampler.py can be used inside IPython to interactively see results from the models being trained.

See my blog post at blog.otoro.net for more details on training on the MNIST set.

This version is an experimental hacked version of the MNIST model to train on CIFAR-10. Results are not really good yet, but I decided to just put the code up in case anyone wants to play with it and make it work.

I wrote another blog post about some of the current generative results on CIFAR-10, and what I think can improve this model going forward.

Tested this implementation on TensorFlow 0.60.

Used images2gif.py written by Almar Klein, Ant1, Marius van Voorden.

License

BSD - images2gif.py

MIT - everything else