biuyq / CT-GAN

48 stars 16 forks source link

CT-GAN

This project is for reproducing the results of paper improving the improved wgans. To test the generative model, please change to the tensorflow_generative_model forder (Tensorflow version: 1.2.1). To test the semi-supervised learning results please change to the Theano_classifier forder (Theano version:0.9.0).

1. For the generative model in MNIST dataset with 1000 data, please run: CT_gan_mnist.py.

For the generative model in CIFAR-10 with 1000 data, run: CT_gan_cifar.py.

For the resnet version of CT_GAN, please run: CT_gan_cifar_resnet.py.

To check some samples generated by our method, please refer to the samples forder.

2. For the semi-supervised classification work, run: CT-MNIST.py for 100 labeled MNIST training.

Run: CT_CIFAR.py OR CT_CIFAR-10_TE.py for 4000 labeled CIFAR-10 training.

The version of time saving one is CT-CIFAR-10_TE.py with temporal ensembling and the regular one is CT_CIFAR.py.