SOTAVerified

Triple Generative Adversarial Networks

2019-12-20Code Available0· sign in to hype

Chongxuan Li, Kun Xu, Jiashuo Liu, Jun Zhu, Bo Zhang

Code Available — Be the first to reproduce this paper.

Reproduce

Code

Abstract

We propose a unified game-theoretical framework to perform classification and conditional image generation given limited supervision. It is formulated as a three-player minimax game consisting of a generator, a classifier and a discriminator, and therefore is referred to as Triple Generative Adversarial Network (Triple-GAN). The generator and the classifier characterize the conditional distributions between images and labels to perform conditional generation and classification, respectively. The discriminator solely focuses on identifying fake image-label pairs. Under a nonparametric assumption, we prove the unique equilibrium of the game is that the distributions characterized by the generator and the classifier converge to the data distribution. As a byproduct of the three-player mechanism, Triple-GAN is flexible to incorporate different semi-supervised classifiers and GAN architectures. We evaluate Triple-GAN in two challenging settings, namely, semi-supervised learning and the extreme low data regime. In both settings, Triple-GAN can achieve excellent classification results and generate meaningful samples in a specific class simultaneously. In particular, using a commonly adopted 13-layer CNN classifier, Triple-GAN outperforms extensive semi-supervised learning methods substantially on more than 10 benchmarks no matter data augmentation is applied or not.

Tasks

Benchmark Results

DatasetModelMetricClaimedVerifiedStatus
CIFAR-10, 1000 LabelsTriple-GAN-V2 (CNN-13, no aug)Accuracy81.81Unverified
CIFAR-10, 1000 LabelsTriple-GAN-V2 (CNN-13)Accuracy85Unverified
CIFAR-10, 1000 LabelsTriple-GAN-V2 (ResNet-26)Accuracy91.59Unverified
CIFAR-10, 4000 LabelsTriple-GAN-V2 (ResNet-26)Percentage error6.54Unverified
CIFAR-10, 4000 LabelsTriple-GAN-V2 (CNN-13)Percentage error10.01Unverified
CIFAR-10, 4000 LabelsTriple-GAN-V2 (CNN-13, no aug)Percentage error12.41Unverified
SVHN, 1000 labelsTriple-GAN-V2 (CNN-13)Accuracy96.55Unverified
SVHN, 1000 labelsTriple-GAN-V2 (CNN-13, no aug)Accuracy96.04Unverified
SVHN, 250 LabelsTriple-GAN-V2 (CNN-13)Accuracy96.52Unverified
SVHN, 250 LabelsTriple-GAN-V2 (CNN-13, no aug)Accuracy95.81Unverified
SVHN, 500 LabelsTriple-GAN-V2 (CNN-13, no aug)Accuracy96.16Unverified
SVHN, 500 LabelsTriple-GAN-V2 (CNN-13)Accuracy96.39Unverified

Reproductions