OmniVerifier-M1: Multimodal Meta-Verifier with Explicit Structured Recalibration

2026-05-27

AI Summary

Current general-purpose foundation models lack reliable, fine-grained multimodal verification mechanisms, hindering their safe and controllable deployment. This work proposes OmniVerifier-M1, a multimodal meta-verification framework that introduces symbolic verification evidence—such as bounding boxes—in place of purely textual explanations. The framework employs a decoupled reinforcement learning strategy to separately optimize binary judgment and meta-verification objectives, integrating rule-driven reward mechanisms with multimodal large language models. This approach substantially improves verification accuracy and interpretability, enables region-level error localization and dynamic self-correction, and further gives rise to the M1-TTS proxy generation system, thereby enhancing the reliability and controllability of multimodal models.

Abstract

Visual outcomes are increasingly central to multimodal large language models, making reliable and fine-grained verification essential for scaling generalist foundation models. In this work, we investigate multimodal meta-verification, which leverages verifier-generated rationales rather than decision-only signals, and explore how to effectively incorporate meta-verification feedback into multimodal verifier training. We identify two key findings. First, symbolic verifier outputs (e.g., bounding boxes) outperform textual explanations as meta-verification rationales, enabling efficient rule-based reinforcement learning rewards while avoiding reliance on model-based rewards from auxiliary judge models. Second, decoupling reinforcement learning objectives for binary judgment and meta-verification substantially outperforms joint reward optimization, due to intrinsic differences in output structure and learning dynamics. Based on these insights, we train OmniVerifier-M1, a generalist visual verifier leveraging symbolic meta-verification and decoupled reinforcement learning. OmniVerifier-M1 provides robust verification and fine-grained error localization, and further enables M1-TTS, a verifier-driven agentic generation system achieving dynamic region-level self-correction. This approach paves the way for more reliable, interpretable, and fine-grained multimodal verification, supporting safer and more controllable foundation model deployment.

Problem

• multimodal verification
• visual outcomes
• meta-verification
• fine-grained error localization
• foundation models

Innovation

• multimodal meta-verification
• symbolic rationales
• decoupled reinforcement learning
• error localization
• agentic self-correction