Learning Disentangled Prompts for Compositional Image Synthesis

TL;DR

This paper introduces a method for domain-adaptive image synthesis that uses visual prompt tuning to learn disentangled prompts for semantic and style features, enabling synthesis of novel images with minimal data.

Contribution

It proposes a novel source class distilled visual prompt framework that learns disentangled semantic and domain prompts from few images for improved compositional image synthesis.

Findings

Effective synthesis of images in new styles with few images

Improved zero-shot domain adaptation classification accuracy

Qualitative results demonstrating compositional generalization

Abstract

We study domain-adaptive image synthesis, the problem of teaching pretrained image generative models a new style or concept from as few as one image to synthesize novel images, to better understand the compositional image synthesis. We present a framework that leverages a pretrained class-conditional generation model and visual prompt tuning. Specifically, we propose a novel source class distilled visual prompt that learns disentangled prompts of semantic (e.g., class) and domain (e.g., style) from a few images. Learned domain prompt is then used to synthesize images of any classes in the style of target domain. We conduct studies on various target domains with the number of images ranging from one to a few to many, and show qualitative results which show the compositional generalization of our method. Moreover, we show that our method can help improve zero-shot domain adaptation classification accuracy.