MaGNET: Uniform Sampling from Deep Generative Network Manifolds Without Retraining

Deep Generative Networks (DGNs) are extensively employed in Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs), and their variants to approximate the data manifold and distribution. However, training samples are often distributed in a non-uniform fashion on the manifold, due to costs or convenience of collection. For example, the CelebA dataset contains a large fraction of smiling faces. These inconsistencies will be reproduced when sampling from the trained DGN, which is not always preferred, e.g., for fairness or data augmentation. In response, we develop MaGNET, a novel and theoretically motivated latent space sampler for any pre-trained DGN, that produces samples uniformly distributed on the learned manifold. We perform a range of experiments on various datasets and DGNs, e.g., for the state-of-the-art StyleGAN2 trained on FFHQ dataset, uniform sampling via MaGNET increases distribution precision and recall by 4.1% & 3.0% and decreases gender bias by 41.2%, without requiring labels or retraining. As uniform distribution does not imply uniform semantic distribution, we also explore separately how semantic attributes of generated samples vary under MaGNET sampling. Figure 1: Random batches of StyleGAN2 (ψ = 0.5) samples with 1024 × 1024 resolution, generated using standard sampling (left), uniform sampling via MaGNET on the learned pixel-space manifold (middle), and uniform sampling on the style-space manifold (right) of the same model. MaGNET sampling yields a higher number of young faces, better gender balance, and greater background/accessory variation, without the need for labels or retraining. Images are sorted by gender-age and color coded red-green (female-male) according to Microsoft Cognitive API predictions. Larger batches of images and attribute distributions are furnished in Appendix E.