Curriculum Direct Preference Optimization for Diffusion and Consistency Models

TL;DR

This paper introduces Curriculum DPO, a two-stage training method for text-to-image models that uses curriculum learning to improve alignment and aesthetics, outperforming existing fine-tuning approaches.

Contribution

It presents a novel curriculum learning framework for DPO, incorporating difficulty-based sampling of training pairs to enhance text-to-image generation quality.

Findings

Outperforms state-of-the-art fine-tuning methods on nine benchmarks.

Improves text alignment, aesthetics, and human preference.

Uses rank difference as a measure of training pair difficulty.

Abstract

Direct Preference Optimization (DPO) has been proposed as an effective and efficient alternative to reinforcement learning from human feedback (RLHF). In this paper, we propose a novel and enhanced version of DPO based on curriculum learning for text-to-image generation. Our method is divided into two training stages. First, a ranking of the examples generated for each prompt is obtained by employing a reward model. Then, increasingly difficult pairs of examples are sampled and provided to a text-to-image generative (diffusion or consistency) model. Generated samples that are far apart in the ranking are considered to form easy pairs, while those that are close in the ranking form hard pairs. In other words, we use the rank difference between samples as a measure of difficulty. The sampled pairs are split into batches according to their difficulty levels, which are gradually used to train the generative model. Our approach, Curriculum DPO, is compared against state-of-the-art fine-tuning approaches on nine benchmarks, outperforming the competing methods in terms of text alignment, aesthetics and human preference. Our code is available at https://github.com/CroitoruAlin/Curriculum-DPO.