In order to evaluate visual localization and place recognition performance over longer periods of time, we provide benchmark datasets aimed at evaluating 6 degree-of-freedom pose estimation accuracy over large appearance variations caused by changes in lighting conditions, seasons, snow, and day-night changes. Each dataset consists of a set of reference images, together with their corresponding ground truth poses. A triangulated 3D model is also be provided for each dataset. For detailed information about the datasets, please see our CVPR 2018 paper.
We have now made available two of the three datasets, the RobotCar-Seasons dataset and the CMU-Seasons dataset. We are currently working on getting the Aachen Day-Night dataset up online as soon as possible.
We are also working on setting up a benchmarking server for automatic evaluation of your localization results. If you are working on a publication and would like to evaluate your method, please see the instructions for evaluation in the readme for the corresponding dataset.