FA-GAN: Feature-Aware GAN for Text to Image Synthesis

TL;DR

FA-GAN introduces a feature-aware GAN framework with a self-supervised discriminator and feature loss, significantly improving the quality of text-to-image synthesis by producing clearer textures and intact objects.

Contribution

The paper proposes a novel FA-GAN model that integrates a self-supervised discriminator and feature-aware loss to enhance image quality in text-to-image synthesis.

Findings

Achieved a lower FID score of 24.58 on MS-COCO, indicating higher image quality.

Outperformed previous state-of-the-art methods in text-to-image synthesis.

Demonstrated improved object integrity and texture clarity in generated images.

Abstract

Text-to-image synthesis aims to generate a photo-realistic image from a given natural language description. Previous works have made significant progress with Generative Adversarial Networks (GANs). Nonetheless, it is still hard to generate intact objects or clear textures (Fig 1). To address this issue, we propose Feature-Aware Generative Adversarial Network (FA-GAN) to synthesize a high-quality image by integrating two techniques: a self-supervised discriminator and a feature-aware loss. First, we design a self-supervised discriminator with an auxiliary decoder so that the discriminator can extract better representation. Secondly, we introduce a feature-aware loss to provide the generator more direct supervision by employing the feature representation from the self-supervised discriminator. Experiments on the MS-COCO dataset show that our proposed method significantly advances the state-of-the-art FID score from 28.92 to 24.58.