UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control

doi:arXiv:2502.05749

	UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control
	Zhu, Kaizhen1,2 ; Pan, Mokai1,2 ; Ma, Yuexin1,2 ; Fu, Yanwei 3; Yu, Jingyi1,2 ; Wang, Jingya1,2 ; Shi, Ye1,2
	2025-02-11
状态	已发表
摘要	Recent advances in diffusion bridge models leverage Doob's $h$-transform to establish fixed endpoints between distributions, demonstrating promising results in image translation and restoration tasks. However, these approaches frequently produce blurred or excessively smoothed image details and lack a comprehensive theoretical foundation to explain these shortcomings. To address these limitations, we propose UniDB, a unified framework for diffusion bridges based on Stochastic Optimal Control (SOC). UniDB formulates the problem through an SOC-based optimization and derives a closed-form solution for the optimal controller, thereby unifying and generalizing existing diffusion bridge models. We demonstrate that existing diffusion bridges employing Doob's $h$-transform constitute a special case of our framework, emerging when the terminal penalty coefficient in the SOC cost function tends to infinity. By incorporating a tunable terminal penalty coefficient, UniDB achieves an optimal balance between control costs and terminal penalties, substantially improving detail preservation and output quality. Notably, UniDB seamlessly integrates with existing diffusion bridge models, requiring only minimal code modifications. Extensive experiments across diverse image restoration tasks validate the superiority and adaptability of the proposed framework. Our code is available at https://github.com/UniDB-SOC/UniDB/.
语种	英语
DOI	arXiv:2502.05749
相关网址	查看原文
出处	Arxiv
收录类别	PPRN.PPRN
WOS记录号	PPRN:121299166
WOS类目	Computer Science, Artificial Intelligence ; Computer Science, Software Engineering ; Engineering, Electrical& Electronic
资助项目	ShanghaiTech AI4S Initiative[SHTAI4S202404] ; Shanghai Local College Capacity Building Program[23010503100] ; NSFC[62303319]
文献类型	预印本
条目标识符	https://kms.shanghaitech.edu.cn/handle/2MSLDSTB/507013
专题	信息科学与技术学院_PI研究组_汪婧雅组信息科学与技术学院_PI研究组_虞晶怡组信息科学与技术学院_硕士生信息科学与技术学院_本科生信息科学与技术学院_PI研究组_马月昕信息科学与技术学院_PI研究组_石野组
通讯作者	Zhu, Kaizhen
作者单位	1.ShanghaiTech Univ, Shanghai, Peoples R China 2.Human Machine Collaborat, MoE Key Lab Intelligent Percept, Shanghai, Peoples R China 3.Fudan Univ, Shanghai, Peoples R China
推荐引用方式 GB/T 7714	Zhu, Kaizhen,Pan, Mokai,Ma, Yuexin,et al. UniDB: A Unified Diffusion Bridge Framework via Stochastic Optimal Control. 2025.