Multi-Step Visual Reasoning with Visual Tokens Scaling and Verification

TL;DR

This paper introduces a dynamic, verifier-guided visual reasoning framework for multi-modal large language models, enabling iterative refinement of visual understanding during inference, which improves accuracy and interpretability.

Contribution

It proposes a novel inference-time visual token scaling method with verifier-guided reasoning, formulated as a Markov Decision Process, and introduces a new dataset for training and evaluation.

Findings

Outperforms existing methods on visual reasoning benchmarks

Enables iterative, context-aware visual reasoning during inference

Provides more interpretable and grounded reasoning processes

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, which is 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.