Now that chainer official implementation is updated. Please refer it.

chainer-VAE

implementation of https://github.com/pfnet/chainer/tree/master/examples/vae using Trainer.

result/loss.png was compiled by using Viz.js. If you use homebrew and wanna dump image file, brew install graphviz, then dot -T <jpeg, png , etc> <computational_graph.dot> -o <output_image>.

beta-VAE

As you can see in http://openreview.net/forum?id=Sy2fzU9gl, you can impose stronger regularization on latent space when define model via net.VAE(C=beta), where, beta > 1.

Execute train_vae.py

Argument Parser

• resume: resume from snapshot. Basically, snapshots are chainer-VAE/result/snapshot_epoch_{epoch}
• interval: save images & snapshot intervals, default is 5