Skip to content
No description or website provided.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


This is the code for semi-supervised learning experiments described in the paper 'Global versus Localized Generative Adversarial Nets' [pdf].

The code is modified from the repository of 'Improved Techniques for Training GANs'

Current status: Initial release

Required Libraries:

  • Theano
  • Lasagne
  • gpuarray
  1. Semi-supervised Learning on Cifar-10

Please download all the files to your dictionary first. To conduct the semi-supervised learning on Cifar-10, please run the following commands:

THEANO_FLAGS='device=<cuda>,floatX=float32' python [--batch_size <100>|--count <400>|...]

To accelerate the training process, LGAN can be trained in two phases2. The first one is the training with only non-Jacobian related parameters by

THEANO_FLAGS='device=<cuda>,floatX=float32' python [--batch_size <100>|--count <400>|--save_dir <./model>|...]

Then followed by the training with Jacobian related parameters by

THEANO_FLAGS='device=<cuda>,floatX=float32' python [--batch_size <100>|--count <400>|--phase1_model_dir <./model>|...]
  1. Semi-supervised Learning on SVHN

Coming soon...

You can’t perform that action at this time.