Code for the models in "PixelVAE: A Latent Variable Model for Natural Images" (https://arxiv.org/abs/1611.05013
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
lib initial commit Jan 16, 2017
tflib
.gitignore
README.md Update README.md Jan 17, 2017
mnist_pixelvae_evaluate.py Update mnist_pixelvae_evaluate.py Jan 16, 2017
mnist_pixelvae_train.py
pixelvae.py 64px big onelevel Jun 29, 2017
read_output.py initial commit Jan 16, 2017

README.md

PixelVAE

Code for the models in PixelVAE: A Latent Variable Model for Natural Images

Binarized MNIST

To train:

python models/mnist_pixelvae_train.py -L 12 -fs 5 -algo cond_z_bias -dpx 16 -ldim 16

To evaluate, take the weights of the model with best validation score from the above training procedure and then run

python models/mnist_pixelvae_evaluate.py -L 12 -fs 5 -algo cond_z_bias -dpx 16 -ldim 16 -w path/to/weights.pkl

Real-valued MNIST, LSUN Bedrooms, 64x64 ImageNet

To train, evaluate, and generate samples:

python pixelvae.py

By default, this runs on real-valued MNIST. You can specify different datasets or model settings within pixelvae.py.