Hal-Eval: A Universal and Fine-grained Hallucination Evaluation Framework for Large Vision Language Models

TL;DR

Hal-Eval introduces a comprehensive framework for evaluating hallucinations in large vision-language models, including a new event hallucination category, using advanced LLMs to generate detailed assessment data.

Contribution

The paper presents a new taxonomy of hallucinations, especially event hallucinations, and develops a universal evaluation framework combining discriminative and generative methods for LVLMs.

Findings

New taxonomy including event hallucinations

Generation of fine-grained hallucination data using LLMs

A comprehensive benchmark for LVLM hallucination assessment

Abstract

Large Vision Language Models exhibit remarkable capabilities but struggle with hallucinations inconsistencies between images and their descriptions. Previous hallucination evaluation studies on LVLMs have identified hallucinations in terms of objects, attributes, and relations but overlooked complex hallucinations that create an entire narrative around a fictional entity. In this paper, we introduce a refined taxonomy of hallucinations, featuring a new category: Event Hallucination. We then utilize advanced LLMs to generate and filter fine grained hallucinatory data consisting of various types of hallucinations, with a particular focus on event hallucinations, laying the groundwork for integrating discriminative and generative evaluation methods within our universal evaluation framework. The proposed benchmark distinctively assesses LVLMs ability to tackle a broad spectrum of hallucinations, making it a reliable and comprehensive tool for gauging LVLMs efficacy in handling hallucinations. We will release our code and data.