The Retrieval-SfM dataset is used for instance-based image retrieval. The dataset contains 28,559 images from 713 locations around the world. Each image has an annotation indicating which location it belongs to.

Most of the locations are famous buildings, such as palaces and towers, which are relatively static and help in recognizing visual locations. The training set contains a variety of variations, including changes in viewpoint, occlusion, and lighting conditions.