From Pixels to Tokens: Revisiting Object Hallucinations in Large Vision-Language Models

2024-10-09

arXiv.org

AI Summary

Large Vision-Language Models (LVLMs) suffer from object hallucination—generating objects absent in the input image—thereby undermining factual consistency and reliability. This work identifies insufficient visual feature disentanglement during modality alignment—not deficient visual encoder representation—as the primary cause. To address this, we propose PATCH tuning: a plug-and-play, architecture-agnostic fine-tuning strategy that introduces bounding-box-guided adaptive virtual tokens for fine-grained, spatially localizable visual feature extraction; integrates modular feature disentanglement alignment; and performs end-to-end multimodal fine-tuning. Crucially, PATCH requires no modification to the backbone architecture. Evaluated across multiple multimodal hallucination benchmarks, PATCH achieves state-of-the-art performance, significantly reducing out-of-image object generation while improving factual consistency and model reliability.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Hallucinations in large vision-language models (LVLMs) are a significant challenge, i.e., generating objects that are not presented in the visual input, which impairs their reliability. Recent studies often attribute hallucinations to a lack of understanding of visual input, yet ignore a more fundamental issue: the model's inability to effectively extract or decouple visual features. In this paper, we revisit the hallucinations in LVLMs from an architectural perspective, investigating whether the primary cause lies in the visual encoder (feature extraction) or the modal alignment module (feature decoupling). Motivated by our findings on the preliminary investigation, we propose a novel tuning strategy, PATCH, to mitigate hallucinations in LVLMs. This plug-and-play method can be integrated into various LVLMs, utilizing adaptive virtual tokens to extract object features from bounding boxes, thereby addressing hallucinations caused by insufficient decoupling of visual features. PATCH achieves state-of-the-art performance on multiple multi-modal hallucination datasets. We hope this approach provides researchers with deeper insights into the underlying causes of hallucinations in LVLMs, fostering further advancements and innovation in this field.

Problem

• Addresses object hallucinations in vision-language models
• Investigates visual feature extraction and decoupling issues
• Proposes plug-and-play method to mitigate false object generation

Innovation

• Plug-and-play tuning with adaptive virtual tokens
• Extracts object features from bounding boxes
• Addresses insufficient decoupling of visual features