VLADBench
收藏资源简介:
VLADBench是一个面向自动驾驶领域的挑战性和细粒度数据集,包含close-form问答,涵盖从静态基础知识到动态路面情况的先进推理。该数据集由华为诺亚方舟实验室和科大联合创建,包含5个关键领域:交通知识理解、通用元素识别、交通图生成、目标属性理解和自我决策与规划。数据集从12个公开数据源精心挑选和构建而成,旨在挑战VLM在多种复杂驾驶情境下的能力。
VLADBench is a challenging and fine-grained dataset tailored for the autonomous driving domain, which includes closed-form question answering tasks covering advanced reasoning spanning from static foundational knowledge to dynamic road surface scenarios. Co-developed by Huawei Noah's Ark Lab and the University of Science and Technology of China (USTC), the dataset encompasses five core research areas: Traffic Knowledge Understanding, General Element Recognition, Traffic Map Generation, Target Attribute Understanding, and Self-Decision-Making and Planning. It is meticulously curated and constructed from 12 public data sources, with the primary goal of challenging the capabilities of Vision-Language Models (VLMs) across various complex driving scenarios.




