EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling

TL;DR

EQ-VAE introduces a regularization technique that enforces equivariance in the latent space of autoencoders, improving the efficiency and quality of image generation without sacrificing reconstruction accuracy.

Contribution

The paper proposes EQ-VAE, a novel regularization method that enforces equivariance in autoencoder latent spaces, enhancing generative model performance and speed.

Findings

Achieves 7x speedup in generative models like DiT-XL/2.

Enhances performance of models such as DiT, SiT, REPA, and MaskGIT.

Compatible with both continuous and discrete autoencoders.

Abstract

Latent generative models have emerged as a leading approach for high-quality image synthesis. These models rely on an autoencoder to compress images into a latent space, followed by a generative model to learn the latent distribution. We identify that existing autoencoders lack equivariance to semantic-preserving transformations like scaling and rotation, resulting in complex latent spaces that hinder generative performance. To address this, we propose EQ-VAE, a simple regularization approach that enforces equivariance in the latent space, reducing its complexity without degrading reconstruction quality. By finetuning pre-trained autoencoders with EQ-VAE, we enhance the performance of several state-of-the-art generative models, including DiT, SiT, REPA and MaskGIT, achieving a 7 speedup on DiT-XL/2 with only five epochs of SD-VAE fine-tuning. EQ-VAE is compatible with both continuous and discrete autoencoders, thus offering a versatile enhancement for a wide range of latent generative models. Project page and code: https://eq-vae.github.io/.