CrowdDriven

CrowdDriven数据集是由查尔姆斯理工大学和Facebook等机构合作创建的户外视觉定位挑战数据集。该数据集包含约3000张图像，通过众包方式收集，覆盖多种地理环境和相机设备。数据集特别设计用于测试和改进在季节、天气、光照和人为变化等条件下的视觉定位算法。CrowdDriven数据集的创建过程中，依赖于人工标注来获取参考姿态，并提供了生成数据集所需的工具，以便于进行有效的二维对应标注。该数据集的应用领域包括自动驾驶汽车、增强/虚拟现实系统等，旨在解决复杂环境下的视觉定位问题。

The CrowdDriven dataset is an outdoor visual localization challenge dataset co-developed by institutions including Chalmers University of Technology and Facebook. It contains approximately 3,000 images collected through crowdsourcing, covering a wide range of geographic environments and camera setups. This dataset is specifically designed to test and optimize visual localization algorithms under various challenging conditions such as seasonal variations, weather changes, lighting fluctuations, and human-induced alterations. During the creation of the CrowdDriven dataset, reference poses were obtained via manual annotation, and tools for dataset generation are provided to facilitate effective 2D correspondence annotation. Its potential application scenarios include autonomous vehicles, augmented/virtual reality systems, and other fields, with the goal of resolving visual localization issues in complex real-world environments.