DT2I: Dense Text-to-Image Generation from Region Descriptions

TL;DR

This paper introduces dense text-to-image synthesis, enabling more intuitive image generation from detailed region descriptions, by proposing a new task and a novel DTC-GAN model that matches image regions with semantic text.

Contribution

The paper proposes the new dense text-to-image synthesis task and introduces DTC-GAN, a novel model with a multi-modal region feature matching loss for improved scene generation.

Findings

Generated plausible images of complex scenes from region captions

DTC-GAN effectively matches image regions with semantic descriptions

Demonstrates advancement in flexible, detailed image synthesis

Abstract

Despite astonishing progress, generating realistic images of complex scenes remains a challenging problem. Recently, layout-to-image synthesis approaches have attracted much interest by conditioning the generator on a list of bounding boxes and corresponding class labels. However, previous approaches are very restrictive because the set of labels is fixed a priori. Meanwhile, text-to-image synthesis methods have substantially improved and provide a flexible way for conditional image generation. In this work, we introduce dense text-to-image (DT2I) synthesis as a new task to pave the way toward more intuitive image generation. Furthermore, we propose DTC-GAN, a novel method to generate images from semantically rich region descriptions, and a multi-modal region feature matching loss to encourage semantic image-text matching. Our results demonstrate the capability of our approach to generate plausible images of complex scenes using region captions.