CoDeGAN: Contrastive Disentanglement for Generative Adversarial Network

TL;DR

CoDeGAN introduces a contrastive approach to improve disentanglement in GANs by relaxing similarity constraints and incorporating self-supervised pre-training, leading to more stable training and better interpretability.

Contribution

It proposes a novel contrastive disentanglement method for GANs that enhances stability and interpretability, integrating self-supervised pre-training for improved unsupervised disentanglement.

Findings

Outperforms state-of-the-art methods on multiple benchmarks

Improves GAN training stability

Enhances disentanglement capabilities

Abstract

Disentanglement, a critical concern in interpretable machine learning, has also garnered significant attention from the computer vision community. Many existing GAN-based class disentanglement (unsupervised) approaches, such as InfoGAN and its variants, primarily aim to maximize the mutual information (MI) between the generated image and its latent codes. However, this focus may lead to a tendency for the network to generate highly similar images when presented with the same latent class factor, potentially resulting in mode collapse or mode dropping. To alleviate this problem, we propose \texttt{CoDeGAN} (Contrastive Disentanglement for Generative Adversarial Networks), where we relax similarity constraints for disentanglement from the image domain to the feature domain. This modification not only enhances the stability of GAN training but also improves their disentangling capabilities. Moreover, we integrate self-supervised pre-training into CoDeGAN to learn semantic representations, significantly facilitating unsupervised disentanglement. Extensive experimental results demonstrate the superiority of our method over state-of-the-art approaches across multiple benchmarks. The code is available at https://github.com/learninginvision/CoDeGAN.