Adaptive Weighted Discriminator for Training Generative Adversarial Networks

CVPR 2021  ·  Vasily Zadorozhnyy, Qiang Cheng, Qiang Ye ·

Generative adversarial network (GAN) has become one of the most important neural network models for classical unsupervised machine learning. A variety of discriminator loss functions have been developed to train GAN's discriminators and they all have a common structure: a sum of real and fake losses that only depends on the actual and generated data respectively. One challenge associated with an equally weighted sum of two losses is that the training may benefit one loss but harm the other, which we show causes instability and mode collapse. In this paper, we introduce a new family of discriminator loss functions that adopts a weighted sum of real and fake parts, which we call adaptive weighted loss functions or aw-loss functions. Using the gradients of the real and fake parts of the loss, we can adaptively choose weights to train a discriminator in the direction that benefits the GAN's stability. Our method can be potentially applied to any discriminator model with a loss that is a sum of the real and fake parts. Experiments validated the effectiveness of our loss functions on an unconditional image generation task, improving the baseline results by a significant margin on CIFAR-10, STL-10, and CIFAR-100 datasets in Inception Scores and FID.

PDF Abstract CVPR 2021 PDF CVPR 2021 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Conditional Image Generation CIFAR-10 aw-BigGAN Inception score 9.52 # 5
FID 6.89 # 7
Image Generation CIFAR-10 aw-AutoGAN Inception score 9.01 # 28
FID 11.82 # 98
Image Generation CIFAR-10 aw-SN-GAN Inception score 8.53 # 39
FID 12.32 # 102
Conditional Image Generation CIFAR-10 aw-SN-GAN Inception score 9 # 7
FID 8.03 # 9
Conditional Image Generation CIFAR-100 aw-BigGAN Inception Score 11.22 # 2
FID 10.23 # 4
Conditional Image Generation CIFAR-100 aw-SN-GAN Inception Score 9.48 # 4
FID 14 # 5
Image Generation CIFAR-100 aw-SN-GAN FID 19.08 # 5
Inception Score 8.31 # 2
Image Generation CIFAR-100 aw-AutoGAN FID 19 # 4
Inception Score 8.9 # 1
Image Generation STL-10 aw-AutoGAN FID 26.32 # 14
Inception score 9.59 # 9
Image Generation STL-10 aw-SN-GAN FID 34.72 # 17
Inception score 9.61 # 8

Methods