Transferring Unconditional to Conditional GANs with Hyper-Modulation

TL;DR

This paper introduces a method for transferring high-quality unconditional GANs to conditional GANs using hyper-modulation, self-initialization, and contrastive learning, enabling efficient adaptation with limited data.

Contribution

It proposes a novel hyper-modulated network architecture with self-initialization and contrastive learning for effective transfer from unconditional to conditional GANs.

Findings

Hyper-modulated networks improve transfer efficiency.

Self-initialization prevents overfitting and mode collapse.

Contrastive learning enhances sample efficiency with small batches.

Abstract

GANs have matured in recent years and are able to generate high-resolution, realistic images. However, the computational resources and the data required for the training of high-quality GANs are enormous, and the study of transfer learning of these models is therefore an urgent topic. Many of the available high-quality pretrained GANs are unconditional (like StyleGAN). For many applications, however, conditional GANs are preferable, because they provide more control over the generation process, despite often suffering more training difficulties. Therefore, in this paper, we focus on transferring from high-quality pretrained unconditional GANs to conditional GANs. This requires architectural adaptation of the pretrained GAN to perform the conditioning. To this end, we propose hyper-modulated generative networks that allow for shared and complementary supervision. To prevent the additional weights of the hypernetwork to overfit, with subsequent mode collapse on small target domains, we introduce a self-initialization procedure that does not require any real data to initialize the hypernetwork parameters. To further improve the sample efficiency of the transfer, we apply contrastive learning in the discriminator, which effectively works on very limited batch sizes. In extensive experiments, we validate the efficiency of the hypernetworks, self-initialization and contrastive loss for knowledge transfer on standard benchmarks.