Bidirectional Conditional Generative Adversarial Networks

TL;DR

This paper introduces BiCoGAN, a bidirectional conditional GAN that disentangles latent variables and auxiliary information, enabling more accurate encoding and generation of conditioned data samples.

Contribution

The paper proposes BiCoGAN, a novel bidirectional cGAN with an effective training method and improved disentanglement of latent and auxiliary variables.

Findings

BiCoGAN encodes auxiliary information more accurately.

BiCoGAN generates more disentangled and effective conditioned samples.

The training involves a novel extrinsic factor loss with dynamic importance tuning.

Abstract

Conditional Generative Adversarial Networks (cGANs) are generative models that can produce data samples ($x$) conditioned on both latent variables ($z$) and known auxiliary information ($c$). We propose the Bidirectional cGAN (BiCoGAN), which effectively disentangles $z$ and $c$ in the generation process and provides an encoder that learns inverse mappings from $x$ to both $z$ and $c$, trained jointly with the generator and the discriminator. We present crucial techniques for training BiCoGANs, which involve an extrinsic factor loss along with an associated dynamically-tuned importance weight. As compared to other encoder-based cGANs, BiCoGANs encode $c$ more accurately, and utilize $z$ and $c$ more effectively and in a more disentangled way to generate samples.