利用视觉监督缓解视觉语言模型中的幻觉
Mitigating Hallucination in Visual Language Models with Visual Supervision
AI总结：
本文通过引入详细视觉标注、SAM与掩码预测损失辅助监督来缓解LVLMs幻觉，并提出RAH-Bench基准，在LLaVA上取得+8.4%的提升。
摘要：
大型视觉语言模型（LVLMs）经常遭受幻觉问题，偶尔会生成与图像内容明显矛盾的响应。关键问题在于其在多模态上下文中理解细节内容的能力较弱，这主要可归因于训练数据和损失函数两个因素。视觉指令数据集主要关注全局描述，而自回归损失函数更偏向文本建模而非图像理解。在本文中，我们引入更详细的视觉标注和更具判别力的视觉模型来促进LVLMs的训练，使其能够生成更精确的响应而不会遭遇幻觉。一方面，我们在全景场景图数据集（PSG）中生成带有详细关系标注的图像-文本对。这些对话更加关注图像中的详细事实，鼓励模型基于多模态上下文回答问题。另一方面，我们集成SAM和掩码预测损失作为辅助监督，迫使LVLMs具备识别与上下文相关对象的能力，从而生成更准确的响应，缓解幻觉。此外，为了对LVLMs中的幻觉进行更深入的评估，我们提出了一个新的基准，RAH-Bench。它将视觉幻觉分为三种不同类型，分别以错误的类别、属性或关系与图像相矛盾，并为每种类型引入假阳性率作为详细的子指标。在该基准上，我们的方法与原始LLaVA相比实现了+8.4%的提升，并在其他模型上取得了广泛的性能改进。
英文摘要：
Large vision-language models (LVLMs) suffer from hallucination a lot, generating responses that apparently contradict to the image content occasionally. The key problem lies in its weak ability to comprehend detailed content in a multi-modal context, which can be mainly attributed to two factors in training data and loss function. The vision instruction dataset primarily focuses on global description, and the auto-regressive loss function favors text modeling rather than image understanding. In this paper, we bring more detailed vision annotations and more discriminative vision models to facilitate the training of LVLMs, so that they can generate more precise responses without encounter hallucination. On one hand, we generate image-text pairs with detailed relationship annotations in panoptic scene graph dataset (PSG). These conversations pay more attention on detailed facts in the image, encouraging the model to answer questions based on multi-modal contexts. On the other hand, we integrate SAM and mask prediction loss as auxiliary supervision, forcing the LVLMs to have the capacity to identify context-related objects, so that they can generate more accurate responses, mitigating hallucination. Moreover, to provide a deeper evaluation on the hallucination in LVLMs, we propose a new benchmark, RAH-Bench. It divides vision hallucination into three different types that contradicts the image with wrong categories, attributes or relations, and introduces False Positive Rate as detailed sub-metric for each type. In this benchmark, our approach demonstrates an +8.4% enhancement compared to original LLaVA and achieves widespread performance improvements across other models.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
