ReasonUni：推理增强重建推进统一多模态模型

Authors:Feize Wu, Zhiyuan Yan, Yuhuan Hong, Yuwei Niu, Zongjian Li, Zheyuan Liu, Bin Lin, Jinbo Yan, Jie Qin, Xudong Mao, Limeng Qiao, Guanglu Wan, Li Yuan

View PDF HTML (experimental)

Abstract:Image-to-text (I2T) understanding and text-to-image (T2I) generation are dual tasks in unified multimodal models (UMMs). Reconstructive post-training couples these tasks, but existing reconstruction-based methods remain perception-centric and provide limited supervision for reasoning about structure, intent, and transformation. We propose ReasonUni, a reasoning-augmented reconstructive post-training framework that extends this coupling beyond perception. Given an original image and an editing instruction, the UMM first reasons about what and how to edit and describes the target state, then generates the corresponding target image conditioned on this understanding output. To jointly optimize both modules, we introduce TreeGRPO, a tree-structured reinforcement learning method that samples multiple reasoning paths and candidate images per path, then aggregates visual rewards to update both modules under a unified objective. Experiments across multiple benchmarks demonstrate consistent gains in both understanding and generation, showing that reasoning-augmented reconstruction yields genuine mutual gains for UMMs.

Comments:	27 pages, including appendix
Subjects:	Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as:	arXiv:2610.20781 [cs.CV]
	(or arXiv:2610.20781v1 [cs.CV] for this version)
	https://doi.org/10.48550/arXiv.2610.20781 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Feize Wu [view email]
[v1] Wed, 14 Oct 2026 18:27:53 UTC (14,528 KB)