Exploring Sparse MoE in GANs for Text-conditioned Image Synthesis

TL;DR

Aurora introduces a sparse Mixture-of-Experts GAN for text-to-image synthesis, achieving competitive zero-shot FID scores at low resolution by adaptively selecting experts based on text and global context.

Contribution

The paper presents Aurora, a novel sparse MoE GAN architecture that improves text-conditioned image synthesis efficiency and quality, with adaptive expert routing based on text and global features.

Findings

Achieves 6.2 zero-shot FID on MS COCO at 64x64 resolution.

Employs adaptive expert routing conditioned on text and global latent code.

Demonstrates the effectiveness of sparse MoE in large-scale text-to-image GANs.

Abstract

Due to the difficulty in scaling up, generative adversarial networks (GANs) seem to be falling from grace on the task of text-conditioned image synthesis. Sparsely-activated mixture-of-experts (MoE) has recently been demonstrated as a valid solution to training large-scale models with limited computational resources. Inspired by such a philosophy, we present Aurora, a GAN-based text-to-image generator that employs a collection of experts to learn feature processing, together with a sparse router to help select the most suitable expert for each feature point. To faithfully decode the sampling stochasticity and the text condition to the final synthesis, our router adaptively makes its decision by taking into account the text-integrated global latent code. At 64x64 image resolution, our model trained on LAION2B-en and COYO-700M achieves 6.2 zero-shot FID on MS COCO. We release the code and checkpoints to facilitate the community for further development.