Streetlearn Intra-City Geolocation Dataset

We provide a complete set of SIFT keypoint descriptors (4,096 keypoints per image) for the complete Streetlearn dataset coverage of Manhattan and Pittsburgh. This highly densely sampled Streetlearn dataset of imagery was used to develop an accurate inter-city predictor and then experimentally resolve, for the first time, the intra-city performance limits of framing image geolocation as a regression-type problem. Also included are mapping files relating each of the numerically labeled imaged to the original Streetlearn panorama id of each city. Finally, we provide zipped archives containing the annually separated imagery considered for our temporal model experiment. Only a sample of 1000 descriptors for each of Manhattan and Pittsburgh are provided here. The complete dataset of descriptors (~59GB in total) is available upon reasonable request.

本研究为覆盖曼哈顿与匹兹堡的完整Streetlearn数据集，提供全套SIFT（Scale-Invariant Feature Transform）关键点描述子，每张图像对应4096个关键点。该高密度采样的Streetlearn图像数据集被用于构建精准的跨城市预测模型，并首次通过实验明确了将图像地理定位建模为回归任务时的城市内性能边界。同时附带映射文件，可将每张带数字编号的图像关联至对应城市的原始Streetlearn全景图像ID。最后，本研究还提供了经年度拆分的压缩图像归档文件，用于时序模型实验。本次仅提供曼哈顿与匹兹堡各1000个描述子的样本集。完整的描述子数据集总容量约59GB，可通过合理申请获取。