This dataset is made publicly available with the paper "TextPlace: Visual Place Recognition and Topological Localization through Reading Scene Texts ", by Ziyang Hong, Yvan Petillot, David Lane, Yishu Miao and Sen Wang, published in the ICCV 2019. This dataset is manually-annotated ground-truth for sequences specifically captured for visual place recognition tasks in urban scenes. The recordings were made with a side-looking camera. Specifically this dataset poses significant challenges in viewpoint, illumination changes and dynamic objects.
Download the dataset: https://drive.google.com/file/d/1KF5ARs6fVQun9zigx3byJRr0fAF49Xfp/view?usp=sharing