Home

Awesome

Semi-Supervised Learning with Sobolev GAN

Code accompanying the papers:

Tested for python 2.7, PyTorch 0.3.0.

To reproduce the CIFAR-10 result for 4000 labeled samples using the K+1 critic, imposing Fisher constraint on full critic f and Sobolev constraint on f- only:

python main.py --dataset cifar10 --dataroot ${DATAROOT} --cuda --outputdir ${OUTPUTDIR} \
--labeledSamples 4000 \
--SSL_critic_type Kp1 --f_component_Fisher f --f_component_Sobolev fneg