Enregistré dans:
Détails bibliographiques
Auteurs principaux: Wen, Zimo, Li, Boxiu, Zhang, Wanbo, Lei, Junxiang, Chen, Xiaoyu, Fan, Yijia, Zhang, Qi, Wang, Yujiang, Qiu, Lili, Li, Bo, Liu, Ziwei, Shan, Caihua, Yang, Yifan, Shen, Yifei
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:https://arxiv.org/abs/2603.03241
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866908863242960896
author Wen, Zimo
Li, Boxiu
Zhang, Wanbo
Lei, Junxiang
Chen, Xiaoyu
Fan, Yijia
Zhang, Qi
Wang, Yujiang
Qiu, Lili
Li, Bo
Liu, Ziwei
Shan, Caihua
Yang, Yifan
Shen, Yifei
author_facet Wen, Zimo
Li, Boxiu
Zhang, Wanbo
Lei, Junxiang
Chen, Xiaoyu
Fan, Yijia
Zhang, Qi
Wang, Yujiang
Qiu, Lili
Li, Bo
Liu, Ziwei
Shan, Caihua
Yang, Yifan
Shen, Yifei
contents Unified multimodal models have recently demonstrated strong generative capabilities, yet whether and when generation improves understanding remains unclear. Existing benchmarks lack a systematic exploration of the specific tasks where generation facilitates understanding. To this end, we introduce UniG2U-Bench, a comprehensive benchmark categorizing generation-to-understanding (G2U) evaluation into 7 regimes and 30 subtasks, requiring varying degrees of implicit or explicit visual transformations. Extensive evaluation of over 30 models reveals three core findings: 1) Unified models generally underperform their base Vision-Language Models (VLMs), and Generate-then-Answer (GtA) inference typically degrades performance relative to direct inference. 2) Consistent enhancements emerge in spatial intelligence, visual illusions, or multi-round reasoning subtasks, where enhanced spatial and shape perception, as well as multi-step intermediate image states, prove beneficial. 3) Tasks with similar reasoning structures and models sharing architectures exhibit correlated behaviors, suggesting that generation-understanding coupling induces class-consistent inductive biases over tasks, pretraining data, and model architectures. These findings highlight the necessity for more diverse training data and novel paradigms to fully unlock the potential of unified multimodal modeling.
format Preprint
id arxiv_https___arxiv_org_abs_2603_03241
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle UniG2U-Bench: Do Unified Models Advance Multimodal Understanding?
Wen, Zimo
Li, Boxiu
Zhang, Wanbo
Lei, Junxiang
Chen, Xiaoyu
Fan, Yijia
Zhang, Qi
Wang, Yujiang
Qiu, Lili
Li, Bo
Liu, Ziwei
Shan, Caihua
Yang, Yifan
Shen, Yifei
Computer Vision and Pattern Recognition
Artificial Intelligence
Unified multimodal models have recently demonstrated strong generative capabilities, yet whether and when generation improves understanding remains unclear. Existing benchmarks lack a systematic exploration of the specific tasks where generation facilitates understanding. To this end, we introduce UniG2U-Bench, a comprehensive benchmark categorizing generation-to-understanding (G2U) evaluation into 7 regimes and 30 subtasks, requiring varying degrees of implicit or explicit visual transformations. Extensive evaluation of over 30 models reveals three core findings: 1) Unified models generally underperform their base Vision-Language Models (VLMs), and Generate-then-Answer (GtA) inference typically degrades performance relative to direct inference. 2) Consistent enhancements emerge in spatial intelligence, visual illusions, or multi-round reasoning subtasks, where enhanced spatial and shape perception, as well as multi-step intermediate image states, prove beneficial. 3) Tasks with similar reasoning structures and models sharing architectures exhibit correlated behaviors, suggesting that generation-understanding coupling induces class-consistent inductive biases over tasks, pretraining data, and model architectures. These findings highlight the necessity for more diverse training data and novel paradigms to fully unlock the potential of unified multimodal modeling.
title UniG2U-Bench: Do Unified Models Advance Multimodal Understanding?
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2603.03241