Soft-IntroVAE: Analyzing and Improving the Introspective Variational Autoencoder
This work improves the stability and theoretical understanding of the IntroVAE model, benefiting researchers and practitioners working with generative models for image synthesis and related applications.
The original IntroVAE model, despite its strong image generation capabilities, suffered from training instability due to its hinge-loss formulation. This paper introduces Soft-IntroVAE, which replaces the hinge-loss with a smooth exponential loss, significantly improving training stability and enabling a complete theoretical analysis. The improved model achieves competitive image generation and reconstruction, and demonstrates compelling results in unsupervised image translation and out-of-distribution detection.
The recently introduced introspective variational autoencoder (IntroVAE) exhibits outstanding image generations, and allows for amortized inference using an image encoder. The main idea in IntroVAE is to train a VAE adversarially, using the VAE encoder to discriminate between generated and real data samples. However, the original IntroVAE loss function relied on a particular hinge-loss formulation that is very hard to stabilize in practice, and its theoretical convergence analysis ignored important terms in the loss. In this work, we take a step towards better understanding of the IntroVAE model, its practical implementation, and its applications. We propose the Soft-IntroVAE, a modified IntroVAE that replaces the hinge-loss terms with a smooth exponential loss on generated samples. This change significantly improves training stability, and also enables theoretical analysis of the complete algorithm. Interestingly, we show that the IntroVAE converges to a distribution that minimizes a sum of KL distance from the data distribution and an entropy term. We discuss the implications of this result, and demonstrate that it induces competitive image generation and reconstruction. Finally, we describe two applications of Soft-IntroVAE to unsupervised image translation and out-of-distribution detection, and demonstrate compelling results. Code and additional information is available on the project website -- https://taldatech.github.io/soft-intro-vae-web