Large vision language models reduce image hallucinations with fine visual experts
PLANE: Plug-and-Play Fine-Grained Visual Experts for Hallucination Mitigation in Large Vision-Language Models
Computer Vision and Pattern RecognitionArtificial IntelligenceComputation and Language
Summary
Large vision-language models sometimes make mistakes by imagining objects or details that aren’t in images. The authors found that part of the problem is the way these models look mostly at main objects and miss smaller details. To fix this, they added special fine-grained image analysis parts that focus on detailed aspects like objects, their attributes, and relationships. They also created a new test to better measure these image mistakes. Their method reduces those hallucinations while keeping the models fast.
What this means in practice
- •For vision-language model developers: Improve model accuracy by integrating fine-grained visual experts to reduce hallucinated objects and incorrect attributes in image descriptions.
- •For ai system integrators: Deploy hallucination mitigation modules to enhance reliability of visual understanding in AI applications involving image captioning and question answering.
Authors
Jianghao Yin, Lang Yu, Qianxi Yu, Qin Chen, Jie Zhou, Zhikai Lei, Liang He
Abstract
Large Vision-Language Models (LVLMs) commonly suffer from hallucinations, which appear as fabricating non-existent objects or outputting incorrect attributes and spatial relationships in images. Previous works empirically indicate that CLIP-style vision encoders tend to focus on primary objects while ignoring image details, which is one of the causes of hallucinations. To address this issue, we propose plug-and-play fine-grained visual experts (PLANE) comprising two components: LoRA visual experts operating on CLIP for multi-aspect image encoding, and an instruction-aware gate network for adaptive image feature fusion. During training, we construct a fine-grained image-caption pairs dataset, where images are described from three aspects: objects, attributes, and relations between objects. Image-caption pairs from different aspects are used to train the three fine-grained LoRA experts respectively. Aiming at comprehensively evaluate our method's capabilities, we manually annotate a fine-grained hallucination evaluation benchmark for LVLMs which is GPT4-independent during evaluation phase (GIFT). We conduct experiments both on popular hallucination evaluation benchmarks and GIFT. The results show that our method effectively reduces hallucinations across models of different types and sizes while introducing minimal inference latency. Compared to advanced baselines, our model also demonstrates superior performance.