Boost Your Human Image Generation Model via Direct Preference Optimization

TL;DR

This paper introduces HG-DPO, an improved direct preference optimization method that uses real images and curriculum learning to enhance the realism and personalization of human image generation models.

Contribution

The paper proposes HG-DPO, a novel DPO framework incorporating real images and curriculum learning to significantly improve human image synthesis quality and personalization.

Findings

Enhanced realism in generated human images.

Effective personalization for identity-specific image generation.

Improved training stability and convergence.

Abstract

Human image generation is a key focus in image synthesis due to its broad applications, but even slight inaccuracies in anatomy, pose, or details can compromise realism. To address these challenges, we explore Direct Preference Optimization (DPO), which trains models to generate preferred (winning) images while diverging from non-preferred (losing) ones. However, conventional DPO methods use generated images as winning images, limiting realism. To overcome this limitation, we propose an enhanced DPO approach that incorporates high-quality real images as winning images, encouraging outputs to resemble real images rather than generated ones. However, implementing this concept is not a trivial task. Therefore, our approach, HG-DPO (Human image Generation through DPO), employs a novel curriculum learning framework that gradually improves the output of the model toward greater realism, making training more feasible. Furthermore, HG-DPO effectively adapts to personalized text-to-image tasks, generating high-quality and identity-specific images, which highlights the practical value of our approach.