Multi-Step Visual Reasoning with Visual Tokens Scaling and Verification

Tianyi Bai1,3,*, Zengjie Hu2,*, Fupeng Sun4,*, Qiu Jiantao3, \(^\ddagger\),
Yizhen Jiang2, Guangxin He1, Bohan Zeng2, Conghui He3,†, Binhang Yuan1,†, Wentao Zhang2,†
1HKUST, 2Peking University, 3Shanghai AI Lab, 4Imperial College London
*Equal contribution. †Project leader. \(^\ddagger\)Corresponding authors.
wentao.zhang@pku.edu.cn, biyuan@ust.hk, {qiujiantao, heconghui}@pjlab.org.cn
arXiv Code Model Dataset

Abstract

Multi-modal large language models (MLLMs) have achieved remarkable capabilities by integrating visual perception with language understanding, enabling applications such as image-grounded dialogue, visual question answering, and scientific analysis. However, most MLLMs adopt a static inference paradigm, encoding the entire image into fixed visual tokens upfront, which limits their ability to iteratively refine understanding or adapt to context during inference. This contrasts sharply with human perception, which is dynamic, selective, and feedback-driven. In this work, we introduce a novel framework for inference-time visual token scaling that enables MLLMs to perform iterative, verifier-guided reasoning over visual content. We formulate the problem as a Markov Decision Process, involving a reasoner that proposes visual actions and a verifier—trained via multi-step Direct Preference Optimization (DPO)—that evaluates these actions and determines when reasoning should terminate. To support this, we present a new dataset, VTS, comprising supervised reasoning trajectories (VTS-SFT) and preference-labeled reasoning comparisons (VTS-DPO). Our method significantly outperforms existing approaches across diverse visual reasoning benchmarks, offering not only improved accuracy but also more interpretable and grounded reasoning processes. These results demonstrate the promise of dynamic inference mechanisms for enabling fine-grained, context-aware visual reasoning in next-generation MLLMs.

VTS-V Framework

Method

Iterative Visual Reasoning with VTS-V.

Our framework equips both open-source and closed-source models with dynamic visual token scaling and step-wise verification to solve complex visual tasks. The example shows how VTS-V: (1) decomposes questions into executable steps, (2) invokes vision tools, and (3) iteratively refines answers via verifier feedback, achieving correct results. In contrast, vanilla models fail to ground detailed visual operations without token scaling, leading to incorrect answers.

Data Construction Pipeline

Pipeline for Synthetic Data Generation and Curation in VTS-V

Our data construction process consists of three stages: (1) generating multi-step reasoning trajectories with visual tool calls, (2) filtering out incorrect trajectories using an LLM-as-a-judge framework, and (3) creating contrastive (correct vs. incorrect) trajectory pairs for multi-step DPO training.

Results

BLINK Benchmark Performance

BLINK Results

General Benchmark Performance

General Benchmarks

Citation


        @article{bai2025vtsv,
          title={Multi-Step Visual Reasoning with Visual Tokens Scaling and Verification},
          author={Bai, Tianyi and Hu, Zengjie and Sun, Fupeng and Qiu, Jiantao and Jiang, Yizhen and He, Guangxin and Zeng, Bohan and He, Conghui and Yuan, Binhang and Zhang, Wentao},
          journal={arXiv preprint arXiv:2506.07235},
          year={2025},
          url={https://arxiv.org/abs/2506.07235},
          archivePrefix={arXiv},
          eprint={2506.07235},
          primaryClass={cs.CV},
        }