MLLM-as-a-Judge: Assessing Multimodal LLM-as-a-Judge with Vision-Language Benchmark

2024/02/07 by Dongping Chen, Ruoxi Chen, Chen, Dongping +17 · 82 citations
Arts and Humanities · Computer Science · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Natural Language Processing Techniques #Translation Studies and Practices #linguistics and terminology studies

paper · pdf · doi:10.48550/arxiv.2402.04788

openalex publication_date 2024/02/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Multimodal Large Language Models (MLLMs) have gained significant attention recently, showing remarkable potential in artificial general intelligence. However, assessing the utility of MLLMs presents considerable challenges, primarily due to the absence of multimodal benchmarks that align with human preferences. Drawing inspiration from the concept of LLM-as-a-Judge within LLMs, this paper introduces a novel benchmark, termed MLLM-as-a-Judge, to assess the ability of MLLMs in assisting judges across diverse modalities, encompassing three distinct tasks: Scoring Evaluation, Pair Comparison, and Batch Ranking. Our study reveals that, while MLLMs demonstrate remarkable human-like discernment in Pair Comparison, there is a significant divergence from human preferences in Scoring Evaluation and Batch Ranking. Furthermore, a closer examination reveals persistent challenges in the judgment capacities of LLMs, including diverse biases, hallucinatory responses, and inconsistencies in judgment, even in advanced models such as GPT-4V. These findings emphasize the pressing need for enhancements and further research efforts to be undertaken before regarding MLLMs as fully reliable evaluators. In light of this, we advocate for additional efforts dedicated to supporting the continuous development within the domain of MLLM functioning as judges. The code and dataset are publicly available at our project homepage: \urlhttps://mllm-judge.github.io/.

Cited by

• Reconstruction as a Bridge for Event-Based Visual Question Answering
• TriDF: Evaluating Perception, Detection, and Hallucination for Interpretable DeepFake Detection
• Becoming Experienced Judges: Selective Test-Time Learning for Evaluators
• CAuSE: Decoding Multimodal Classifiers using Faithful Natural Language Explanation
• ProAgent: Harnessing On-Demand Sensory Contexts for Proactive LLM Agent Systems
• EditThinker: Unlocking Iterative Reasoning for Any Image Editor
• Self-Improving VLM Judges Without Human Annotations
• PPTArena: A Benchmark for PowerPoint Editing
• LLM2Fx-Tools: Tool Calling For Music Post-Production
• A Progressive Evaluation Framework for Multicultural Analysis of Story Visualization
• Multi-Crit: Benchmarking Multimodal Judges on Pluralistic Criteria-Following
• DualVLA: Building a Generalizable Embodied Agent via Partial Decoupling of Reasoning and Action
• EWE: An Agentic Framework for Extreme Weather Analysis
• WaymoQA: A Multi-View Visual Question Answering Dataset for Safety-Critical Reasoning in Autonomous Driving
• CREward: A Type-Specific Creativity Reward Model
• Musical Score Understanding Benchmark: Evaluating Large Language Models' Comprehension of Complete Musical Scores
• Beyond Multiple Choice: Verifiable OpenQA for Robust Vision-Language RFT
• Multi-Agent Multimodal Large Language Model Framework for Automated Interpretation of Fuel Efficiency Analytics in Public Transportation
• MedGEN-Bench: Contextually entangled benchmark for open-ended multimodal medical generation
• MM-CRITIC: A Holistic Evaluation of Large Multimodal Models as Multimodal Critique
• Multimodal Large Language Models for Low-Resource Languages: A Case Study for Basque
• MACEval: A Multi-Agent Continual Evaluation Network for Large Models
• DiagramIR: An Automatic Pipeline for Educational Math Diagram Evaluation
• Automating Hardware Design and Verification from Architectural Papers via a Neural-Symbolic Graph Framework
• Med-Banana: Learning Quality-Controlled Medical Image Editing from Success-and-Failure Trajectories
• LLM-as-a-Judge for Evaluating System Responses in Conversational Music Recommendation
• LLM-as-a-Verifier: A General-Purpose Verification Framework
• SmoothGuard: Defending Multimodal Large Language Models with Noise Perturbation and Clustering Aggregation
• Automatically Benchmarking LLM Code Agents through Agent-Driven Annotation and Evaluation
• Omni-Reward: Towards Generalist Omni-Modal Reward Modeling with Free-Form Preferences
• M³T2IBench: A Large-Scale Multi-Category, Multi-Instance, Multi-Relation Text-to-Image Benchmark
• VEHME: A Vision-Language Model For Evaluating Handwritten Mathematics Expressions
• Paper2Web: Let's Make Your Paper Alive!
• Robust Driving QA through Metadata-Grounded Context and Task-Specific Prompts
• WebDevJudge: Evaluating (M)LLMs as Critiques for Web Development Quality
• Disentanglement Beyond Static vs. Dynamic: A Benchmark and Evaluation Framework for Multi-Factor Sequential Representations
• TREAT: A Code LLMs Trustworthiness / Reliability Evaluation and Testing Framework
• Uniworld-V2: Reinforce Image Editing with Diffusion Negative-aware Finetuning and MLLM Implicit Feedback
• UniME-V2: MLLM-as-a-Judge for Universal Multimodal Embedding Learning
• Multi-Agent Debate for LLM Judges with Adaptive Stability Detection
• Judge's Verdict: A Comprehensive Analysis of LLM Judge Capability Through Human Agreement
• AutoPR: Let's Automate Your Academic Promotion!
• MDSEval: A Meta-Evaluation Benchmark for Multimodal Dialogue Summarization
• CIR-CoT: Towards Interpretable Composed Image Retrieval via End-to-End Chain-of-Thought Reasoning
• Deploying Tiny LVLM Judges for Real-World Evaluation of Chart Models: Lessons Learned and Best Practices
• BrowserArena: Evaluating LLM Agents on Real-World Web Navigation Tasks
• OptAgent: Optimizing Query Rewriting for E-commerce via Multi-Agent Simulation
• Automated Evaluation can Distinguish the Good and Bad AI Responses to Patient Questions about Hospitalization
• Automated Model Discovery via Multi-modal & Multi-step Pipeline
• WoW: Towards a World omniscient World model Through Embodied Interaction
• Unlocking the Essence of Beauty: Advanced Aesthetic Reasoning with Relative-Absolute Policy Optimization
• VideoJudge: Bootstrapping Enables Scalable Supervision of MLLM-as-a-Judge for Video Understanding
• TrustJudge: Inconsistencies of LLM-as-a-Judge and How to Alleviate Them
• RadAgents: Multimodal Agentic Reasoning for Chest X-ray Interpretation with Radiologist-like Workflows
• BaseReward: A Strong Baseline for Multimodal Reward Model
• MARS2 2025 Challenge on Multimodal Reasoning: Datasets, Methods, Results, Discussion, and Outlook
• Towards Rationale-Answer Alignment of LVLMs via Self-Rationale Calibration
• What Makes a Good Generated Image? Investigating Human and Multimodal LLM Image Preference Alignment
• AesBiasBench: Evaluating Bias and Alignment in Multimodal Language Models for Personalized Image Aesthetic Assessment
• VideoRewardBench: Comprehensive Evaluation of Multimodal Reward Models for Video Understanding
• MM-SeR: Multimodal Self-Refinement for Lightweight Image Captioning
• Med-RewardBench: Benchmarking Reward Models and Judges for Medical Multimodal Large Language Models
• Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning
• A Novel Framework for Automated Explain Vision Model Using Vision-Language Models
• GLSim: Detecting Object Hallucinations in LVLMs via Global-Local Similarity
• EMNLP: Educator-role Moral and Normative Large Language Models Profiling
• Drifting Away from Truth: GenAI-Driven News Diversity Challenges LVLM-Based Misinformation Detection
• TalkPlayData 2: An Agentic Synthetic Data Pipeline for Multimodal Conversational Music Recommendation
• HOID-R1: Reinforcement Learning for Open-World Human-Object Interaction Detection Reasoning with Multimodal Large Language Model
• Generating Dialogues from Egocentric Instructional Videos for Task Assistance: Dataset, Method and Benchmark
• MultiRef: Controllable Image Generation with Multiple Visual References
• LoRA in LoRA: Towards Parameter-Efficient Architecture Expansion for Continual Visual Instruction Tuning
• Towards Robust Evaluation of Visual Activity Recognition: Resolving Verb Ambiguity with Sense Clustering
• ConfProBench: A Confidence Evaluation Benchmark for MLLM-Based Process Judges
• Are We on the Right Way for Assessing Document Retrieval-Augmented Generation?
• LaTCoder: Converting Webpage Design to Code with Layout-as-Thought
• Personalized Safety Alignment for Text-to-Image Diffusion Models
• Sel3DCraft: Interactive Visual Prompts for User-Friendly Text-to-3D Generation
• Evaluating the Efficacy of Large Language Models for Generating Fine-Grained Visual Privacy Policies in Homes
• MLLM-CTBench: A Benchmark for Continual Instruction Tuning with Reasoning Process Diagnosis
• RecGPT Technical Report
• TARS: MinMax Token-Adaptive Preference Strategy for Hallucination Reduction in MLLMs

Related

• MLLM-Bench: Evaluating Multimodal LLMs with Per-sample Criteria
• Judge Anything: MLLM as a Judge Across Any Modality
• From Generation to Judgment: Opportunities and Challenges of LLM-as-a-judge
• MME-Survey: A Comprehensive Survey on Evaluation of Multimodal LLMs
• JudgeLM: Fine-tuned Large Language Models are Scalable Judges