A Cognitive Evaluation Benchmark of Image Reasoning and Description for Large Vision Language Models

2024-02-28

arXiv.org

AI Summary

This work addresses the lack of systematic evaluation of higher-order cognitive capabilities in Large Vision-Language Models (LVLMs). To this end, we introduce CogniBench—the first benchmark specifically designed to assess higher-order cognition in LVLMs. Inspired by the clinical “Cookie Theft” neuropsychological test, CogniBench comprises 251 semantically rich images and pioneers the adaptation of human clinical cognitive assessment paradigms to LVLM evaluation. We formally define eight cognitive dimensions—including abstract reasoning and causal inference—and establish a dual-track, fine-grained annotation framework covering image captioning and visual question answering. Experimental results reveal that state-of-the-art LVLMs achieve accuracies below 45% on abstract reasoning and causal inference tasks—substantially underperforming human baselines—thereby exposing fundamental limitations in their higher-order cognitive reasoning abilities.

Technology Category

• Computer Vision: Language and Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large Vision-Language Models (LVLMs), despite their recent success, are hardly comprehensively tested for their cognitive abilities. Inspired by the prevalent use of the Cookie Theft task in human cognitive tests, we propose a novel evaluation benchmark to evaluate high-level cognitive abilities of LVLMs using images with rich semantics. The benchmark consists of 251 images along with comprehensive annotations. It defines eight reasoning capabilities and comprises an image description task and a visual question answering task. Our evaluation of well-known LVLMs shows that there is still a significant gap in cognitive abilities between LVLMs and humans.

Problem

• Evaluate cognitive abilities of LVLMs
• Propose novel benchmark using rich semantic images
• Identify gap between LVLMs and human cognition

Innovation

• Cognitive evaluation benchmark for LVLMs
• Rich semantic images for testing
• Eight reasoning capabilities defined