AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Multimodal Meta-Verifier with Explicit Structured Recalibration

Xinchen Zhang, Bowei Liu, Jiale Liu, Chufan Shi, Yizhen Zhang, Junhong Liu, Youliang Zhang, Zhiheng Li, Yujiu Yang, Ling Yang

ICML 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Multimodal LLM benchmarks

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Investigates multimodal meta-verification, where verifiers produce rationales rather than decision-only signals, and studies how to incorporate meta-verification feedback into multimodal verifier training.

Problem

Reliable, fine-grained verification of visual outcomes from multimodal LLMs, which is needed for scaling generalist foundation models.

Approach

Uses verifier-generated rationales (meta-verification) instead of decision-only signals, and proposes explicit structured recalibration to fold meta-verification feedback into verifier training.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
decouple binary-judgment and meta-verification rewards instead of joint reward (ablation)Qwen3-VL-8B verifier · ViVerBench Overallaccuracy: 0.671 -> 0.680Improved
decouple binary-judgment and meta-verification rewards instead of joint reward (ablation)OmniVerifier-7B · ViVerBench Overallaccuracy: 0.661 -> 0.668Improved
decouple rewards instead of joint reward (ablation)OmniVerifier-7B · ViVerBench Countingaccuracy: 0.640 -> 0.710Improved
decouple rewards instead of joint reward (ablation)Qwen3-VL-8B · RefCOCO Overallaccuracy: 0.847 -> 0.866Improved
decouple rewards instead of joint reward (ablation)OmniVerifier-7B · ViVerBench Concept Existence Abs.P. (small subtask)accuracy: 0.513 -> 0.506Hurt
decouple rewards instead of joint reward (ablation)Qwen3-VL-8B · ViVerBench Mazeaccuracy: 0.458 -> 0.380Hurt
use rule-based IoU bounding-box meta-reward instead of model-based textual explanation reward (ablation)Qwen3-VL-8B verifier · ViVerBench Overallaccuracy: 0.670 -> 0.671No clear effect
use rule-based IoU bounding-box meta-reward instead of model-based textual explanation reward (ablation)RL verifier training · per-sample reward computationtime (ms): 20.2 -> 0.021Improved
use bounding-box meta-reward instead of textual explanation (ablation)Qwen3-VL-8B · per-card GPU memory (GB)memory: 58.3 -> 49.9Improved
add OmniVerifier-M1 as verifier in M1-TTS on top of RePlanRePlan · T2I-CoreBench Overallscore: 0.629 -> 0.690Improved
add OmniVerifier-M1 verifier to GPT-Image-1.5GPT-Image-1.5 · WISE Overallscore: 0.83 -> 0.88Improved
use OmniVerifier-M1 instead of Qwen3-VL-8B as verifier in M1-TTS with RePlanRePlan · WISE Timescore: 0.64 -> 0.61 on Time; 0.66 -> 0.71 on CulturalMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy