Nullu: Mitigating Object Hallucinations in Large Vision-Language Models via HalluSpace Projection

TL;DR

Nullu is a novel method that reduces object hallucinations in large vision-language models by projecting input features into a null space of hallucinated embeddings, improving output accuracy without extra inference costs.

Contribution

The paper introduces HalluSpace projection and null space orthogonalization as a new approach to mitigate hallucinations in LVLMs, leveraging prior information in LLMs.

Findings

Effectively reduces object hallucinations across LVLMs

Maintains performance on standard benchmarks

No additional inference cost incurred

Abstract

Recent studies have shown that large vision-language models (LVLMs) often suffer from the issue of object hallucinations (OH). To mitigate this issue, we introduce an efficient method that edits the model weights based on an unsafe subspace, which we call HalluSpace in this paper. With truthful and hallucinated text prompts accompanying the visual content as inputs, the HalluSpace can be identified by extracting the hallucinated embedding features and removing the truthful representations in LVLMs. By orthogonalizing the model weights, input features will be projected into the Null space of the HalluSpace to reduce OH, based on which we name our method Nullu. We reveal that HalluSpaces generally contain prior information in the large language models (LLMs) applied to build LVLMs, which have been shown as essential causes of OH in previous studies. Therefore, null space projection suppresses the LLMs' priors to filter out the hallucinated features, resulting in contextually accurate outputs. Experiments show that our method can effectively mitigate OH across different LVLM families without extra inference costs and also show strong performance in general LVLM benchmarks. Code is released at https://github.com/Ziwei-Zheng/Nullu.