IntroVAE: Introspective Variational Autoencoders for Photographic Image Synthesis

TL;DR

IntroVAE is a new generative model that combines variational autoencoders and adversarial training, enabling high-quality photographic image synthesis with stable training and self-evaluation capabilities.

Contribution

The paper introduces IntroVAE, a hybrid model that integrates VAE and GAN principles into a single-stage, discriminator-free architecture for high-resolution image synthesis.

Findings

Produces high-resolution photo-realistic images

Comparable or superior to state-of-the-art GANs

Stable training without extra discriminators

Abstract

We present a novel introspective variational autoencoder (IntroVAE) model for synthesizing high-resolution photographic images. IntroVAE is capable of self-evaluating the quality of its generated samples and improving itself accordingly. Its inference and generator models are jointly trained in an introspective way. On one hand, the generator is required to reconstruct the input images from the noisy outputs of the inference model as normal VAEs. On the other hand, the inference model is encouraged to classify between the generated and real samples while the generator tries to fool it as GANs. These two famous generative frameworks are integrated in a simple yet efficient single-stream architecture that can be trained in a single stage. IntroVAE preserves the advantages of VAEs, such as stable training and nice latent manifold. Unlike most other hybrid models of VAEs and GANs, IntroVAE requires no extra discriminators, because the inference model itself serves as a discriminator to distinguish between the generated and real samples. Experiments demonstrate that our method produces high-resolution photo-realistic images (e.g., CELEBA images at \(1024^{2}\)), which are comparable to or better than the state-of-the-art GANs.