Analyzing and Mitigating Object Hallucination in Large Vision-Language Models

By Yiyang Zhou, Chenhang Cui, Jaehong Yoon, Linjun Zhang, Zhun Deng, Chelsea Finn, Mohit Bansal, Huaxiu Yao. International Conference on Learning Representations, 2023

347 citations recorded by the source corpus.

Large vision-language models (LVLMs) have shown remarkable abilities in understanding visual information with human languages. However, LVLMs still suffer from object hallucination, which is the problem of generating descriptions that include objects that do not actually exist in the images. This can negatively impact many vision-language tasks, such as visual summarization and reasoning. To address this issue, we propose a simple yet powerful algorithm, LVLM Hallucination Revisor (LURE), to post-hoc rectify object hallucination in LVLMs by reconstructing less hallucinatory descriptions. LURE is grounded in a rigorous statistical analysis of the key factors underlying object hallucination, including co-occurrence (the frequent appearance of certain objects alongside others in images), uncertainty (objects with higher uncertainty during LVLM decoding), and object position (hallucination often appears in the later part of the generated text). LURE can also be seamlessly integrated with any LVLMs. We evaluate LURE on six open-source LVLMs, achieving a 23% improvement in general object hallucination evaluation metrics over the previous best approach. In both GPT and human evaluations, LURE consistently ranks at the top. Our data and code are available at https://github.com/YiyangZhou/LURE.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

View Paper: https://www.semanticscholar.org/paper/93c525267e93c78309a5b28a3eb0780704125744

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Zheng Zhang
2. Dahua Lin
3. Xing Sun
4. Mike Zheng Shou
5. Neng H. Yu
6. Sirui Zhao
7. Xiao-wen Dong
8. Tong He

In Computer Vision and Pattern Recognition

1. Dahua Lin
2. Zheng Zhang
3. Luc Van Gool
4. Chunhua Shen
5. Xiaogang Wang
6. Tong He
7. Dacheng Tao
8. Alan L. Yuille

In Radiology, Nuclear Medicine and Imaging

1. Dahua Lin
2. Mike Zheng Shou
3. Tong He
4. Zheng Zhang

Citations in the corpus

• A Survey on Hallucination in Large Vision-Language Models (2024, methodology): “More recently, two post-processing methods were introduced to reduce hallucinations in generated image descriptions: LURE [ Zhou et al. , 2023 ] and Woodpecker [ Yin et al. , 2023b ] .”
• Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding (2023, neutral): “Several recent studies address this issue by proposing hallucination-targeted datasets for fine-tuning [20, 42], training a post-hoc revisor to reconstruct less hallucinatory outputs [77] or adapting factually augmented Reinforcement Learning from Human Feedback (RLHF) [60].”
• OPERA: Alleviating Hallucination in Multi-Modal Large Language Models via Over-Trust Penalty and Retrospection-Allocation (2023, positive): “Various approaches [29, 40, 42, 47] have been proposed to reduce hallucinations in MLLMs.”
• Hallucination of Multimodal Large Language Models: A Survey (2024, positive): “Specifically, inspired by denoising autoencoders [101], which are designed to reconstruct clean data from corrupted input, LURE [137] employs a hallucination revisor that aims to transform potentially hallucinatory descriptions into accurate ones.”
• A survey on multimodal large language models (2023, positive): “More efficient training methods may unlock more power in MLLMs with limited computational resources.”
• Sora: A Review on Background, Technology, Limitations, and Opportunities of Large Vision Models (2024, background): “Concurrently, a significant body of research is focused on devising and implementing methods to reduce hallucinations in these large models [123, 124, 125].”