Causal-HalBench: Uncovering LVLMs Object Hallucinations Through Causal Intervention

TL;DR

This paper introduces Causal-HalBench, a benchmark using causal analysis and counterfactual samples to evaluate and understand object hallucinations caused by spurious correlations in LVLMs.

Contribution

It presents a formal causal framework and a new benchmark for quantifying and analyzing spurious correlations in LVLMs, addressing a gap in current evaluation methods.

Findings

LVLMs are susceptible to spurious correlations leading to hallucinations.

Causal-HalBench effectively quantifies model robustness against spurious correlations.

Mainstream LVLMs show varying degrees of vulnerability to object hallucinations.

Abstract

Large Vision-Language Models (LVLMs) often suffer from object hallucination, making erroneous judgments about the presence of objects in images. We propose this primar- ily stems from spurious correlations arising when models strongly associate highly co-occurring objects during train- ing, leading to hallucinated objects influenced by visual con- text. Current benchmarks mainly focus on hallucination de- tection but lack a formal characterization and quantitative evaluation of spurious correlations in LVLMs. To address this, we introduce causal analysis into the object recognition scenario of LVLMs, establishing a Structural Causal Model (SCM). Utilizing the language of causality, we formally de- fine spurious correlations arising from co-occurrence bias. To quantify the influence induced by these spurious correla- tions, we develop Causal-HalBench, a benchmark specifically constructed with counterfactual samples and integrated with comprehensive causal metrics designed to assess model ro- bustness against spurious correlations. Concurrently, we pro- pose an extensible pipeline for the construction of these coun- terfactual samples, leveraging the capabilities of proprietary LVLMs and Text-to-Image (T2I) models for their genera- tion. Our evaluations on mainstream LVLMs using Causal- HalBench demonstrate these models exhibit susceptibility to spurious correlations, albeit to varying extents.