Semi-supervised infoGAN

Official Torch Code for the paper Guiding InfoGAN with Semi-Supervision.

The script implements the training routine for the MNIST dataset. We used a cropped version of size 28x28 which was taken down. You can find the original MNIST dataset here.

After cloning this repository, download the dataset into the code directory and specify the path to it in the configuration.lua file, alongside any other parameter changes you wish to have. Then execute the training procedure in train_ss-infogan.lua with

th train_ss-infogan.lua

If the score with respect to a classifier is required, a valid classifier must be trained. Its path can then be specified in the configuration.lua file. If none is available, the metric will not be calculated.

Examples

MNIST

SVHN

CelebA

CIFAR-10

Dependencies

luarocks install torchnet
luarocks install penlight
luarocks install nninit
luarocks install cutorch
luarocks install cunn

Credits

This code is a heavily modified version of and based on the code found here.