To evaluate the effect of cross-view image-based geo-localization method under real scenes, we build up our cross-view image dataset. Different with CVUSA or CVACT, our cross-view image dataset is the first cross-view image dataset contains heading data and are collected in real scenarios. It also including different settings such as spatial resolution and north alignment.

The download link of cross-view image pais:https://pan.baidu.com/s/1ftxWnywyZzPdLbcTFSMqXw. Extraction code: cvip

The siamese network model used for image retrieval will be uploaded soon.

The ground-level panoramic images are collected by Ladybug5, Satellite images are collected from google earth and geo-locations of image are collected by GPS receiver of surveying level. The image collection are synchronized with GPS receiver at 1Hz update rate. All the equipment are deployed on Buick GL8.

Our cross-view image pairs are collected in city, campus and rural areas. There are totally 377 pairs for each cross-view image setting.

Satellite images with different spatial resolution but same image width. Spatial resolution from 18 to 20.

The ground-level panoramic images are north aligned or not north aligned. The middle of image are parallel with north for 'north aligned' but not parallel with north for 'not north aligned'

This data set is for academic purpose only, if you have any issues, please contact JunboWang ([email protected])