LanP: Rethinking the Impact of Language Priors in Large Vision-Language Models

TL;DR

This paper introduces LanP, a benchmark to evaluate the strength of language priors in large vision-language models, revealing many models struggle with visual scenarios involving partial object occlusion.

Contribution

The paper proposes LanP, a new benchmark for assessing language priors in LVLMs, and provides extensive experimental analysis across 25 models to understand their limitations.

Findings

Many LVLMs have weak language priors in challenging visual scenarios.

Models like GPT-4 Turbo perform poorly with accuracy below 0.5 on partial occlusion questions.

LanP reveals the need to balance language priors to improve LVLM robustness.

Abstract

Large Vision-Language Models (LVLMs) have shown impressive performance in various tasks. However, LVLMs suffer from hallucination, which hinders their adoption in the real world. Existing studies emphasized that the strong language priors of LVLMs can overpower visual information, causing hallucinations. However, the positive role of language priors is the key to a powerful LVLM. If the language priors are too weak, LVLMs will struggle to leverage rich parameter knowledge and instruction understanding abilities to complete tasks in challenging visual scenarios where visual information alone is insufficient. Therefore, we propose a benchmark called LanP to rethink the impact of Language Priors in LVLMs. It is designed to investigate how strong language priors are in current LVLMs. LanP consists of 170 images and 340 corresponding well-designed questions. Extensive experiments on 25 popular LVLMs reveal that many LVLMs' language priors are not strong enough to effectively aid question answering when objects are partially hidden. Many models, including GPT-4 Turbo, exhibit an accuracy below 0.5 in such a scenario.