AmsterTime
收藏资源简介:
AmsterTime数据集是由代尔夫特理工大学创建,包含2500张精心策划的图像,这些图像匹配了阿姆斯特丹城市的历史档案图像和现代街景图像。数据集通过Mapillary地理信息系统平台直接众包收集,旨在解决视觉地点识别(VPR)中的严重领域转移问题。数据集的创建过程涉及人类专家验证匹配对,确保图像对的准确性。AmsterTime数据集主要用于评估VPR任务中的各种基准模型,包括非学习、监督和自监督方法,特别适用于评估模型在不同视角、光照和外观变化下的性能。
The AmsterTime Dataset was created by Delft University of Technology. It consists of 2500 carefully curated images, which are matched pairs of historical archival images and modern street view images of Amsterdam. The dataset was directly crowdsourced via the Mapillary geographic information system platform, with the aim of addressing the severe domain shift problem in Visual Place Recognition (VPR). During its construction, human experts were engaged to validate all matched image pairs to ensure their accuracy. The AmsterTime Dataset is primarily utilized to evaluate various benchmark models for VPR tasks, including non-learning, supervised, and self-supervised approaches, and it is particularly suitable for assessing model performance under diverse variations in viewpoints, lighting conditions, and scene appearances.




