HaloProbe: Bayesian Detection and Mitigation of Object Hallucinations in Vision-Language Models

TL;DR

HaloProbe is a Bayesian framework that detects and mitigates object hallucinations in vision-language models more effectively and non-invasively than existing methods.

Contribution

It introduces HaloProbe, a novel Bayesian approach that accurately estimates hallucination probabilities by combining internal decoding signals with external description statistics.

Findings

HaloProbe reduces hallucinations more effectively than state-of-the-art methods.

It preserves utility and fluency better than intervention-based mitigation.

External scoring with HaloProbe enables non-invasive hallucination mitigation.

Abstract

Large vision-language models can produce object hallucinations in image descriptions, highlighting the need for effective detection and mitigation strategies. Prior work commonly relies on the model's attention weights on visual tokens as a detection signal. We reveal that coarse-grained attention-based analysis is unreliable due to hidden confounders, specifically token position and object repetition in a description. This leads to Simpson's paradox: the attention trends reverse or disappear when statistics are aggregated. Based on this observation, we introduce HaloProbe, a Bayesian framework that factorizes external description statistics and internal decoding signals to estimate token-level hallucination probabilities. HaloProbe uses balanced training to isolate internal evidence and combines it with a learned prior over external features to recover the true posterior. While intervention-based mitigation methods often degrade utility or fluency by modifying models' internals, we use HaloProbe as an external scoring signal for non-invasive mitigation. Our experiments show that HaloProbe-guided decoding reduces hallucinations more effectively than state-of-the-art intervention-based methods while preserving utility.