Dual Adversarial Inference for Text-to-Image Synthesis

TL;DR

This paper introduces a dual adversarial inference framework for text-to-image synthesis that disentangles content and style in the latent space, leading to improved image quality and meaningful style representations.

Contribution

It proposes a novel dual adversarial inference method to learn disentangled content and style representations in an unsupervised manner for text-to-image synthesis.

Findings

Learned meaningful style representations not described in text

Enhanced image quality on Oxford-102, CUB, and COCO datasets

Unsupervised disentanglement of content and style variables

Abstract

Synthesizing images from a given text description involves engaging two types of information: the content, which includes information explicitly described in the text (e.g., color, composition, etc.), and the style, which is usually not well described in the text (e.g., location, quantity, size, etc.). However, in previous works, it is typically treated as a process of generating images only from the content, i.e., without considering learning meaningful style representations. In this paper, we aim to learn two variables that are disentangled in the latent space, representing content and style respectively. We achieve this by augmenting current text-to-image synthesis frameworks with a dual adversarial inference mechanism. Through extensive experiments, we show that our model learns, in an unsupervised manner, style representations corresponding to certain meaningful information present in the image that are not well described in the text. The new framework also improves the quality of synthesized images when evaluated on Oxford-102, CUB and COCO datasets.