Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Self-Supervised GANs via Auxiliary Rotation Loss

About

Conditional GANs are at the forefront of natural image synthesis. The main drawback of such models is the necessity for labeled data. In this work we exploit two popular unsupervised learning techniques, adversarial training and self-supervision, and take a step towards bridging the gap between conditional and unconditional GANs. In particular, we allow the networks to collaborate on the task of representation learning, while being adversarial with respect to the classic GAN game. The role of self-supervision is to encourage the discriminator to learn meaningful feature representations which are not forgotten during training. We test empirically both the quality of the learned image representations, and the quality of the synthesized images. Under the same conditions, the self-supervised GAN attains a similar performance to state-of-the-art conditional counterparts. Finally, we show that this approach to fully unsupervised learning can be scaled to attain an FID of 23.4 on unconditional ImageNet generation.

Ting Chen, Xiaohua Zhai, Marvin Ritter, Mario Lucic, Neil Houlsby• 2018

Related benchmarks

TaskDatasetResultRank
Image GenerationCIFAR-10 (test)
FID7.52
471
Image ClassificationImageNet (val)
Top-1 Accuracy38.3
354
Image ClassificationImageNet (test)--
235
Unconditional Image GenerationCIFAR-10 (test)
FID15.05
216
Image GenerationCIFAR10 32x32 (test)
FID15.65
154
Image GenerationSTL-10 (test)
Inception Score10.36
59
Image GenerationLSUN bedroom
FID13.3
56
Image GenerationCIFAR-10 32x32
FID15.6
44
Image GenerationTiny ImageNet (test)
IS10.21
35
ClassificationTiny-ImageNet 1.0 (val)
Accuracy34.6
30
Showing 10 of 28 rows

Other info

Code

Follow for update