IllusionVQA: A Challenging Optical Illusion Dataset for Vision Language Models

TL;DR

IllusionVQA introduces a challenging optical illusion dataset to evaluate vision-language models' understanding and reasoning, revealing their limitations in interpreting inherently unreasonable images compared to human performance.

Contribution

This paper presents IllusionVQA, a novel dataset of optical illusions designed to test VLMs' comprehension and localization abilities, highlighting their current weaknesses.

Findings

GPT4V achieves 62.99% accuracy in comprehension

VLMs perform poorly on localization tasks compared to humans

In-Context Learning and Chain-of-Thought reasoning degrade VLM performance

Abstract

The advent of Vision Language Models (VLM) has allowed researchers to investigate the visual understanding of a neural network using natural language. Beyond object classification and detection, VLMs are capable of visual comprehension and common-sense reasoning. This naturally led to the question: How do VLMs respond when the image itself is inherently unreasonable? To this end, we present IllusionVQA: a diverse dataset of challenging optical illusions and hard-to-interpret scenes to test the capability of VLMs in two distinct multiple-choice VQA tasks - comprehension and soft localization. GPT4V, the best performing VLM, achieves 62.99% accuracy (4-shot) on the comprehension task and 49.7% on the localization task (4-shot and Chain-of-Thought). Human evaluation reveals that humans achieve 91.03% and 100% accuracy in comprehension and localization. We discover that In-Context Learning (ICL) and Chain-of-Thought reasoning substantially degrade the performance of Gemini-Pro in the localization task. Tangentially, we discover a potential weakness in the ICL capabilities of VLMs: they fail to locate optical illusions even when the correct answer is in the context window as a few-shot example.