Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zheng, Xuhui, An, Kang, Wang, Ziliang, Wang, Yuhang, Qian, Faqiang, Wu, Yichao
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2512.07203
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911307004903424
author Zheng, Xuhui
An, Kang
Wang, Ziliang
Wang, Yuhang
Qian, Faqiang
Wu, Yichao
author_facet Zheng, Xuhui
An, Kang
Wang, Ziliang
Wang, Yuhang
Qian, Faqiang
Wu, Yichao
contents Multimodal pre-training remains constrained by the descriptive bias of image-caption pairs, leading models to favor surface linguistic cues over grounded visual understanding. We introduce MMRPT, a masked multimodal reinforcement pre-training framework that strengthens visual reasoning in MLLMs. We are the first to incorporate reinforcement learning directly into the pre-training of large vision-language models, enabling learning signals that reward visual grounding rather than caption imitation. MMRPT constructs masked multimodal data by estimating sentence-level visual dependency via attention over visual tokens and masking highly vision-dependent segments; the model reconstructs these spans through vision-grounded reasoning guided by a semantic-visual reward. Experiments show consistent zero-shot gains across diverse benchmarks and substantially improved robustness under supervised fine-tuning, demonstrating that reinforcement-driven masked reasoning provides a more reliable and generalizable pre-training objective for multimodal models.
format Preprint
id arxiv_https___arxiv_org_abs_2512_07203
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MMRPT: MultiModal Reinforcement Pre-Training via Masked Vision-Dependent Reasoning
Zheng, Xuhui
An, Kang
Wang, Ziliang
Wang, Yuhang
Qian, Faqiang
Wu, Yichao
Computer Vision and Pattern Recognition
Multimodal pre-training remains constrained by the descriptive bias of image-caption pairs, leading models to favor surface linguistic cues over grounded visual understanding. We introduce MMRPT, a masked multimodal reinforcement pre-training framework that strengthens visual reasoning in MLLMs. We are the first to incorporate reinforcement learning directly into the pre-training of large vision-language models, enabling learning signals that reward visual grounding rather than caption imitation. MMRPT constructs masked multimodal data by estimating sentence-level visual dependency via attention over visual tokens and masking highly vision-dependent segments; the model reconstructs these spans through vision-grounded reasoning guided by a semantic-visual reward. Experiments show consistent zero-shot gains across diverse benchmarks and substantially improved robustness under supervised fine-tuning, demonstrating that reinforcement-driven masked reasoning provides a more reliable and generalizable pre-training objective for multimodal models.
title MMRPT: MultiModal Reinforcement Pre-Training via Masked Vision-Dependent Reasoning
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.07203