Differentiable Augmentation for Data-Efficient GAN Training

The performance of generative adversarial networks (GANs) heavily deteriorates given a limited amount of training data. This is mainly because the discriminator is memorizing the exact training set... (read more)

PDF Abstract NeurIPS 2020 PDF NeurIPS 2020 Abstract
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Generation CIFAR-10 DiffAugment-BigGAN Inception score 9.16 # 8
FID 8.7 # 10
Image Generation CIFAR-10 DiffAugment-StyleGAN2 Inception score 9.40 # 6
FID 9.89 # 11
Image Generation CIFAR-10 DiffAugment-CR-BigGAN FID 8.49 # 9
Image Generation ImageNet 128x128 DiffAugment-BigGAN FID 6.8 # 3
IS 100.8 # 2

Methods used in the Paper