视觉大模型 / VLM

CARE Drive A Framework for Evaluating Reason-Responsiveness of Vision Language Models in Automated Driving

CARE Drive：一种用于评估视觉语言模型在自动驾驶中推理响应性的框架

Lucas Elbert Suryana, Farah Bierenga, Sanne van Buuren, Pepijn Kooij, Elsefien Tulleners, Federico Scari, Simeon Calvert, Bart van Arem, Arkady Zgonnikov

机构 * organization= Department of Transport \& Planning, Faculty of Civil Engineering ； Geosciences, Delft University of Technology ； organization= Department of Cognitive Robotics, Faculty of Mechanical Engineering, Delft University of Technology ； organization= Centre for Meaningful Human Control, Delft University of Technology ； organization= Faculty of Mechanical Engineering, Delft University of Technology , city= Delft , country= The Netherlands 纠错

专题命中 幻觉与鲁棒性 ：vision language model(title,abstract)；分类 cs.CV、cs.AI

AI总结 CARE Drive提出了一种评估视觉语言模型在自动驾驶中推理响应性的框架，通过受控上下文变化比较基线和增强模型决策，验证人类原因对决策的影响。

Comments 21 pages, on submission to Transportation Research Part C

Hierarchical Refinement of Universal Multimodal Attacks on Vision-Language Models

面向视觉语言模型的层次化通用多模态攻击

Peng-Fei Zhang, Zi Huang

机构 * School of Electrical Engineering and Computer Science, the University of Queensland（电气工程与计算机科学学院，昆士兰大学） ； Department of Computer Science, City University of Hong Kong（计算机科学系，香港城市大学） 纠错

专题命中 幻觉与鲁棒性 ：vision-language model(title)；分类 cs.CV

AI总结 本文提出HRA框架，通过层次化优化路径和文本重要性建模，实现对视觉语言模型的通用多模态攻击，验证了其在多种任务和数据集上的优越迁移性能。

Comments 10 pages, 7 figures

Semantic-Guided 3D Gaussian Splatting for Transient Object Removal

语义引导的3D高斯点云对瞬态物体去除

Aditi Prabakaran, Priyesh Shukla

专题命中 幻觉与鲁棒性 ：vision-language model(abstract)；分类 cs.CV

AI总结 本文提出基于语义分类的3D高斯点云瞬态去除方法，通过CLIP相似度分数和阈值校准提升重建质量，减少内存开销并保持实时渲染性能。