多模态LLM评估

Benchmarks of MLLMs: Survey

A Survey on Benchmarks of Multimodal Large Language Models

¹Tencent, ²PKU, ²NUS, ²SEU, ²NJU

Evaluations and Benchmarks in Context of Multimodal LLM(MLLM Tutorial @ CVPR 2025) ⚡We will actively maintain this repository and incorporate new research as it emerges. If you have any questions, please contact swordli@tencent.com.

📌 What is This Survey About?

<p align="center"> <img src="timeline.jpg" width="100%" height="100%"> </p>

<p align="center"> <img src="BMLLM_statistic.png" width="100%" height="100%"> </p>

Multimodal Large Language Models (MLLMs) are gaining increasing popularity in both academia and industry due to their remarkable performance in various applications such as visual question answering, visual perception, understanding, and reasoning. Over the past few years, significant efforts have been made to examine MLLMs from multiple perspectives. This paper presents a comprehensive review of 200+ benchmarks and evaluations for MLLMs, focusing on (1)perception and understanding, (2)cognition and reasoning, (3)specific domains, (4)key capabilities, and (5)other modalities. Finally, we discuss the limitations of the current evaluation methods for MLLMs and explore promising future directions. Our key argument is that evaluation should be regarded as a crucial discipline to better support the development of MLLMs.

Key Capabilities

#### Conversation Abilities Long-context 1. <mark>Mile-Bench</mark> "MileBench: Benchmarking MLLMs in Long Context". Song D, Chen S, Chen G H, et al.. arXiv 2024. [Paper] [Github]. 2. <mark>MMNeedle</mark> "Multimodal Needle in a Haystack: Benchmarking Long-Context Capability of Multimodal Large Language Models". Wang H, Shi H, Tan S, et al.. arXiv 2024. [Paper] [Github]. 3. <mark>MLVU</mark> "MLVU: A Comprehensive Benchmark for Multi-Task Long Video Understanding". Zhou J, Shu Y, Zhao B, et al.. arXiv 2024. [Paper] [Github]. Instruction Following 1. <mark>CoIN</mark> "CoIN: A Benchmark of Continual Instruction tuNing for Multimodel Large Language Model". Chen C, Zhu J, Luo X, et al.. arXiv 2024. [Paper] [Github]. 2. <mark>MIA-Bench</mark> "MIA-Bench: Towards Better Instruction Following Evaluation of Multimodal LLMs". Qian Y, Ye H, Fauconnier J P, et al.. arXiv 2024. [Paper] [Github]. 3. <mark>DEMON</mark> "Fine-tuning Multimodal LLMs to Follow Zero-shot Demonstrative Instructions". Li J, Pan K, Ge Z, et al.. ICLR 2023. [Paper] [Github]. 4. <mark>VisIT-Bench</mark> "VisIT-Bench: A Benchmark for Vision-Language Instruction Following Inspired by Real-World Use". Bitton Y, Bansal H, Hessel J, et al.. NeurIPS 2023. [Paper] [Github].

#### Hallucination 1. <mark>POPE</mark> "Evaluating Object Hallucination in Large Vision-Language Models". Li Y, Du Y, Zhou K, et al.. EMNLP 2023. [Paper] [Github]. 2. <mark>GAVIE</mark> "Mitigating Hallucination in Large Multi-Modal Models via Robust Instruction Tuning". Liu F, Lin K, Li L, et al.. ICLR 2023. [Paper] [Github]. 3. <mark>HaELM</mark> "Evaluation and Analysis of Hallucination in Large Vision-Language Models". Wang J, Zhou Y, Xu G, et al.. arXiv 2023. [Paper] [Github]. 4. <mark>M-HalDetect</mark> "Detecting and Preventing Hallucinations in Large Vision Language Models". Gunjal A, Yin J, Bas E.. AAAI 2024. [Paper] [Github]. 5. <mark>Bingo</mark> "Holistic Analysis of Hallucination in GPT-4V(ision): Bias and Interference Challenges". Cui C, Zhou Y, Yang X, et al.. arXiv 2023. [Paper] [Github]. 6. <mark>HallusionBench</mark> "HALLUSIONBENCH: An Advanced Diagnostic Suite for Entangled Language Hallucination and Visual Illusion in Large Vision-Language Models". Guan T, Liu F, Wu X, et al.. CVPR 2024. [Paper] [Github]. 7. <mark>VHTest</mark> "Visual Hallucinations of Multi-modal Large Language Models". Huang W, Liu H, Guo M, et al.. arXiv 2024. [Paper] [Github]. 8. <mark>CorrelationQA</mark> "The Instinctive Bias: Spurious Images lead to Hallucination in MLLMs". Han T, Lian Q, Pan R, et al.. arXiv 2024. [Paper] [Github]. 9. <mark>CHAIR</mark> "Object Hallucination in Image Captioning". Rohrbach A, Hendricks L A, Burns K, et al.. EMNLP 2018. [Paper] [Github]. 10. <mark>MHaluBench</mark> "Unified Hallucination Detection for Multimodal Large Language Models". Chen X, Wang C, Xue Y, et al.. arXiv 2024. [Paper] [Github]. 11. <mark>VideoHallucer</mark> "VideoHallucer: Evaluating Intrinsic and Extrinsic Hallucinations in Large Video-Language Models". Wang Y, Wang Y, Zhao D, et al.. arXiv 2024. [Paper] [Github]. 12. <mark>MMHAL-BENCH</mark> "Aligning Large Multimodal Models with Factually Augmented RLHF". Sun Z, Shen S, Cao S, et al.. arXiv 2023. [Paper] [Github]. 13. <mark>AMBER</mark> "AMBER: An LLM-free Multi-dimensional Benchmark for MLLMs Hallucination Evaluation". Wang J, Wang Y, Xu G, et al.. arXiv 2023. [Paper] [Github]. 14. <mark>MMECeption</mark> "GenCeption: Evaluate Multimodal LLMs with Unlabeled Unimodal Data". Cao L, Buchner V, Senane Z, et al.. arXiv 2024. [Paper] [Github].

#### Trustworthiness Robustness 1. <mark>MAD-Bench</mark> "How Easy is It to Fool Your Multimodal LLMs? An Empirical Analysis on Deceptive Prompts". Qian Y, Zhang H, Yang Y, et al.. arXiv 2024. [Paper] [[Github]()]. 2. <mark>MMR</mark> "Seeing Clearly, Answering Incorrectly: A Multimodal Robustness Benchmark for Evaluating MLLMs on Leading Questions". Liu Y, Liang Z, Wang Y, et al.. arXiv 2024. [Paper] [Github]. 3. <mark>MM-SpuBench</mark> "MM-SpuBench: Towards Better Understanding of Spurious Biases in Multimodal LLMs". Ye W, Zheng G, Ma Y, et al.. arXiv 2024. [Paper] [Github]. 4. <mark>MM-SAP</mark> "MM-SAP: A Comprehensive Benchmark for Assessing Self-Awareness of Multimodal Large Language Models in Perception". Wang Y, Liao Y, Liu H, et al.. arXiv 2024. [Paper] [Github]. 5. <mark>BenchLMM</mark> "BenchLMM: Benchmarking Cross-style Visual Capability of Large Multimodal Models". Cai R, Song Z, Guan D, et al.. arXiv 2023. [Paper] [Github]. 6. <mark>VQAv2-IDK</mark> "Visually Dehallucinative Instruction Generation: Know What You Don’t Know". Cha S, Lee J, Lee Y, et al.. ICASSP 2024. [Paper] [Github]. 7. <mark>MVI-Bench</mark> "MVI-Bench: A Comprehensive Benchmark for Evaluating Robustness to Misleading Visual Inputs in LVLMs". Chen H, Peng J, Min D, et al.. ICML 2026. [Paper] [Github].

Safety 1. <mark>MMUBench</mark> "Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models". Li J, Wei Q, Zhang C, et al.. arXiv 2024. [Paper] [[Github]()]. 2. <mark>JailBreakV-28K</mark> "JailBreakV-28K: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks". Luo W, Ma S, Liu X, et al.. arXiv 2024. [Paper] [Github]. 3. <mark>MultiTrust</mark> "Benchmarking Trustworthiness of Multimodal Large Language Models: A Comprehensive Study". Zhang Y, Huang Y, Sun Y, et al.. arXiv 2024. [Paper] [Github]. 4. <mark>MM-SafetyBench</mark> "MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Models". Liu X, Zhu Y, Gu J, et al.. ECCV 2024. [Paper] [Github]. 5. <mark>SHIELD</mark> "SHIELD: An Evaluation Benchmark for Face Spoofing and Forgery Detection with Multimodal Large Language Models". Shi Y, Gao Y, Lai Y, et al.. arXiv 2024. [Paper] [Github]. 6. <mark>RTVLM</mark> "Red teaming visual language models". Li M, Li L, Yin Y, et al.. arXiv 2024. [Paper] [Github].