Photographic Text-to-Image Synthesis with a Hierarchically-nested Adversarial Network

TL;DR

This paper introduces a hierarchical adversarial network for text-to-image synthesis that improves image quality and semantic consistency through novel objectives and architecture, validated on multiple datasets.

Contribution

The paper proposes a hierarchical-nested adversarial training framework with a new generator architecture and a semantic similarity measure for enhanced text-to-image synthesis.

Findings

Significant improvement over previous methods on all datasets.

Enhanced semantic consistency and image fidelity.

Effective use of hierarchical adversarial objectives.

Abstract

This paper presents a novel method to deal with the challenging task of generating photographic images conditioned on semantic image descriptions. Our method introduces accompanying hierarchical-nested adversarial objectives inside the network hierarchies, which regularize mid-level representations and assist generator training to capture the complex image statistics. We present an extensile single-stream generator architecture to better adapt the jointed discriminators and push generated images up to high resolutions. We adopt a multi-purpose adversarial loss to encourage more effective image and text information usage in order to improve the semantic consistency and image fidelity simultaneously. Furthermore, we introduce a new visual-semantic similarity measure to evaluate the semantic consistency of generated images. With extensive experimental validation on three public datasets, our method significantly improves previous state of the arts on all datasets over different evaluation metrics.