Wasserstein Loss for Semantic Editing in the Latent Space of GANs
This addresses the challenge of reliable semantic editing for GAN users, but it is incremental as it offers an alternative to classifier-based methods without major performance gains.
The paper tackled the problem of semantic editing in GAN latent spaces by proposing a Wasserstein loss formulation to avoid issues like out-of-distribution regions and adversarial samples from classifier-based methods, achieving performance on-par with existing approaches on digits and faces datasets using StyleGAN2.
The latent space of GANs contains rich semantics reflecting the training data. Different methods propose to learn edits in latent space corresponding to semantic attributes, thus allowing to modify generated images. Most supervised methods rely on the guidance of classifiers to produce such edits. However, classifiers can lead to out-of-distribution regions and be fooled by adversarial samples. We propose an alternative formulation based on the Wasserstein loss that avoids such problems, while maintaining performance on-par with classifier-based approaches. We demonstrate the effectiveness of our method on two datasets (digits and faces) using StyleGAN2.