There is a widely-spread claim that GANs are difficult to train, and GAN architectures in the literature are littered with empirical tricks. We provide evidence against this claim and build a modern GAN baseline in a more principled manner. First, we derive a well-behaved regularized relativistic GAN loss that addresses issues of mode dropping and non-convergence that were previously tackled via a bag of ad-hoc tricks. We analyze our loss mathematically and prove that it admits local convergence guarantees, unlike most existing relativistic losses. Second, this loss allows us to discard all ad-hoc tricks and replace outdated backbones used in common GANs with modern architectures. Using StyleGAN2 as an example, we present a roadmap of simplification and modernization that results in a new minimalist baseline---R3GAN. Despite being simple, our approach surpasses StyleGAN2 on FFHQ, ImageNet, CIFAR, and Stacked MNIST datasets, and compares favorably against state-of-the-art GANs and diffusion models.

Thenwesynthesizepseudo pair images through the latent space exploration of StyleGAN2 to find implicit associations between different animal faces. Finally, we introduce the semantic matching loss to overcome the problem of extreme shape differences between species.

A large brown metal cube belowa large green rubber cylinder A large gray metal sphereabove a small red metal cube A small red metal cube behinda large brown metal cube A large brown metal cube below a large green rubber cylinder A large gray metal sphereabove a small red metal cube A small red metal cube on the left of a large brown metal cube A large brown metal cube below a large green rubber cylinder A blue objectinfrontofa gray object! A gray object on the left ofa green object A green object behindablue object! A blue objectin front ofa gray object! A gray object behind a green object! A green object on the left ofa blue object! A blue object behind a gray object A gray object on the left ofa green object A green object on the right ofa gray object CLIPQuery imageFine-tuned CLIPOurs( a) Top 1 image-text retrieval result on i Gibsonscenes.(

Generative adversarial networks (GANs) typically require ample data for training in order to synthesize high-fidelity images.