Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lu, Yunhong, Wang, Qichao, Cao, Hengyuan, Xu, Xiaoyin, Zhang, Min
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915330256797696
author Lu, Yunhong
Wang, Qichao
Cao, Hengyuan
Xu, Xiaoyin
Zhang, Min
author_facet Lu, Yunhong
Wang, Qichao
Cao, Hengyuan
Xu, Xiaoyin
Zhang, Min
contents Direct Preference Optimization (DPO) aligns text-to-image (T2I) generation models with human preferences using pairwise preference data. Although substantial resources are expended in collecting and labeling datasets, a critical aspect is often neglected: \textit{preferences vary across individuals and should be represented with more granularity.} To address this, we propose SmPO-Diffusion, a novel method for modeling preference distributions to improve the DPO objective, along with a numerical upper bound estimation for the diffusion optimization objective. First, we introduce a smoothed preference distribution to replace the original binary distribution. We employ a reward model to simulate human preferences and apply preference likelihood averaging to improve the DPO loss, such that the loss function approaches zero when preferences are similar. Furthermore, we utilize an inversion technique to simulate the trajectory preference distribution of the diffusion model, enabling more accurate alignment with the optimization objective. Our approach effectively mitigates issues of excessive optimization and objective misalignment present in existing methods through straightforward modifications. Our SmPO-Diffusion achieves state-of-the-art performance in preference evaluation, outperforming baselines across metrics with lower training costs. The project page is https://jaydenlyh.github.io/SmPO-project-page/.
format Preprint
id arxiv_https___arxiv_org_abs_2506_02698
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences
Lu, Yunhong
Wang, Qichao
Cao, Hengyuan
Xu, Xiaoyin
Zhang, Min
Computer Vision and Pattern Recognition
Direct Preference Optimization (DPO) aligns text-to-image (T2I) generation models with human preferences using pairwise preference data. Although substantial resources are expended in collecting and labeling datasets, a critical aspect is often neglected: \textit{preferences vary across individuals and should be represented with more granularity.} To address this, we propose SmPO-Diffusion, a novel method for modeling preference distributions to improve the DPO objective, along with a numerical upper bound estimation for the diffusion optimization objective. First, we introduce a smoothed preference distribution to replace the original binary distribution. We employ a reward model to simulate human preferences and apply preference likelihood averaging to improve the DPO loss, such that the loss function approaches zero when preferences are similar. Furthermore, we utilize an inversion technique to simulate the trajectory preference distribution of the diffusion model, enabling more accurate alignment with the optimization objective. Our approach effectively mitigates issues of excessive optimization and objective misalignment present in existing methods through straightforward modifications. Our SmPO-Diffusion achieves state-of-the-art performance in preference evaluation, outperforming baselines across metrics with lower training costs. The project page is https://jaydenlyh.github.io/SmPO-project-page/.
title Smoothed Preference Optimization via ReNoise Inversion for Aligning Diffusion Models with Varied Human Preferences
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2506.02698