Improved Techniques for Training GANs

We present a variety of new architectural features and training procedures that we apply to the generative adversarial networks (GANs) framework. We focus on two applications of GANs: semi-supervised learning, and the generation of images that humans find visually realistic. Unlike most work on generative models, our primary goal is not to train a model that assigns high likelihood to test data, nor do we require the model to be able to learn well without using any labels. Using our new techniques, we achieve state-of-the-art results in semi-supervised classification on MNIST, CIFAR-10 and SVHN. The generated images are of high quality as confirmed by a visual Turing test: our model generates MNIST samples that humans cannot distinguish from real data, and CIFAR-10 samples that yield a human error rate of 21.3%. We also present ImageNet samples with unprecedented resolution and show that our methods enable the model to learn recognizable features of ImageNet classes.

PDF Abstract NeurIPS 2016 PDF NeurIPS 2016 Abstract


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Conditional Image Generation CIFAR-10 Improved GAN Inception score 8.09 # 14
Image Generation CIFAR-10 Improved GAN Inception score 6.86 # 59
Semi-Supervised Image Classification CIFAR-10, 4000 Labels GAN Percentage error 15.59 # 37
Image Classification SVHN Improved GAN Percentage error 8.11 # 42

Results from Other Papers

Task Dataset Model Metric Name Metric Value Rank Source Paper Compare
Semi-Supervised Image Classification SVHN, 1000 labels GAN Accuracy 91.89 # 19