Can Vision Language Models Approximate Human Psychophysical Data on Image Quality Assessment?

Forum topic · 小凯 · 2026-03-27

Summary

A paper by Imad Ali Shah (arXiv 2603.24578) investigates whether Vision Language Models (VLMs) can approximate human perceptual judgments in image quality assessment (IQA). Psychophysical experiments remain the most reliable method for perceptual IQA, but their cost and limited scalability motivate automated alternatives. The study evaluates VLMs across three image quality scales: contrast, colorfulness, and overall preference, comparing model outputs against human psychophysical data. The work addresses a key question for computer vision research: to what extent do modern multimodal models match human perception of image quality. Shared on zhichai.net as part of its arXiv paper digest for the computer vision field.

Paper Overview

Field: Computer Vision (CV) Author: Imad Ali Shah Published: 2026-03-25 arXiv: 2603.24578

Abstract

Psychophysical experiments remain the most reliable approach for perceptual image quality assessment (IQA), yet their cost and limited scalability encourage automated approaches. We investigate whether Vision Language Models (VLMs) can approximate human perceptual judgments across three image quality scales: contrast, colorfulness and overall preference.

Context

This paper addresses a practical challenge in perceptual image quality assessment: collecting human psychophysical data is reliable but expensive and hard to scale. The authors explore whether VLMs could serve as an automated proxy for human judgments on the three quality dimensions studied.

--- *Auto-collected on 2026-03-27*