LOGAN: Latent Optimisation for Generative Adversarial Networks
This work addresses training instability in GANs for image generation, offering incremental improvements over existing methods.
The paper tackles the problem of unstable training and mode collapse in generative adversarial networks by introducing latent optimization with natural gradient, which improves adversarial dynamics. The result is state-of-the-art performance on ImageNet (128x128) with an Inception Score of 148 and Fréchet Inception Distance of 3.4, representing improvements of 17% and 32% over the baseline.
Training generative adversarial networks requires balancing of delicate adversarial dynamics. Even with careful tuning, training may diverge or end up in a bad equilibrium with dropped modes. In this work, we improve CS-GAN with natural gradient-based latent optimisation and show that it improves adversarial dynamics by enhancing interactions between the discriminator and the generator. Our experiments demonstrate that latent optimisation can significantly improve GAN training, obtaining state-of-the-art performance for the ImageNet ($128 \times 128$) dataset. Our model achieves an Inception Score (IS) of $148$ and an Fréchet Inception Distance (FID) of $3.4$, an improvement of $17\%$ and $32\%$ in IS and FID respectively, compared with the baseline BigGAN-deep model with the same architecture and number of parameters.