Contrast Sensitivity in Multimodal Large Language Models: A Psychophysics-Inspired Evaluation

2025/08/14 by Pablo Hernández-Cámara, Hernández-Cámara, Pablo, Alexandra Gomez-Villa +9 · 1 citation
Computer Science · #Binary classification #Binary number #Computer Vision and Pattern Recognition (cs.CV) #Contrast (vision) #FOS: Computer and information sciences #Function (biology) #Pattern recognition (psychology) #Perception #Process (computing) #Sensitivity (control systems) #Topic Modeling

paper · open access · doi:10.48550/arxiv.2508.10367

published in PubMed 201, 108903 (National Institutes of Health)

openalex publication_date 2025/08/14 · openalex created_date 2025/10/17 · openalex updated_date 2026/08/05

Abstract

Understanding how Multimodal Large Language Models (MLLMs) process low-level visual features is critical for evaluating their perceptual abilities and has not been systematically characterized. Inspired by human psychophysics, we introduce a behavioural method for estimating the Contrast Sensitivity Function (CSF) in MLLMs by treating them as end-to-end observers. Models are queried with structured prompts while viewing noise-based stimuli filtered at specific spatial frequencies. Psychometric functions are derived from the binary verbal responses. Therefore, contrast thresholds (and CSFs) are obtained without relying on internal activations or classifier-based proxies, as opposed to previous reports on artificial networks. Our results reveal that some models resemble human CSFs in shape or scale, but none capture both. We also find that CSF estimates are highly sensitive to prompt phrasing, indicating limited linguistic robustness. Finally, we show that CSFs predict model performance under frequency-filtered and adversarial conditions. These findings highlight systematic differences in frequency tuning across MLLMs and establish this CSF estimation as a scalable diagnostic tool for multimodal perception.

Citations

• Do computer vision foundation models learn the low-level characteristics of the human visual system?
• SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features
• Qwen2.5-VL Technical Report
• Magma: A Foundation Model for Multimodal AI Agents
• A Turing Test for Artificial Nets devoted to model Human Vision
• The Art of Deception: Color Visual Illusions and Diffusion Models
• Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scaling
• What Did I Do Wrong? Quantifying LLMs' Sensitivity and Consistency to Prompt Engineering
• Are Images Indistinguishable to Humans Also Indistinguishable to Classifiers?
• MindSet: Vision. A toolbox for testing DNNs on key psychological experiments
• Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference
• Spatial-frequency channels, shape bias, and adversarial robustness
• Visual Instruction Tuning
• Learning Transferable Visual Models From Natural Language Supervision
• Contrast sensitivity throughout adulthood
• Can We Talk Models Into Seeing the World Differently?

Cited by

• LLMs as Signal Detectors: Sensitivity, Bias, and the Temperature-Criterion Analogy

Related

• Exploring Multimodal Perception in Large Language Models Through Perceptual Strength Ratings
• The Percept-V Challenge: Can Multimodal LLMs Crack Simple Perception Problems?
• Evaluating and Advancing Multimodal Large Language Models in Perception Ability Lens
• Large language models predict human sensory judgments across six modalities
• Do MLLMs Exhibit Human-like Perceptual Behaviors? HVSBench: A Benchmark for MLLM Alignment with Human Perceptual Behavior