Visual place recognition is the task of recognizing a place depicted in an image based on its pure visual appearance without metadata. In visual place recognition, the challenges lie upon not only the changes in lighting conditions, camera viewpoint, and scale but also the characteristic of scene-level images and the distinct features of the area. To resolve these challenges, one must consider both the local discriminativeness and the global semantic context of images. On the other hand, the diversity of the datasets is also particularly important to develop more general models and advance the progress of the field. In this paper, we present a fully-automated system for place recognition at a city-scale based on content-based image retrieval. Our main contributions to the community lie in three aspects. Firstly, we take a comprehensive analysis of visual place recognition and sketch out the unique challenges of the task compared to general image retrieval tasks. Next, we propose yet a simple pooling approach on top of convolutional neural network activations to embed the spatial information into the image representation vector. Finally, we introduce new datasets for place recognition, which are particularly essential for application-based research. Furthermore, throughout extensive experiments, various issues in both image retrieval and place recognition are analyzed and discussed to give some insights into improving the performance of retrieval models in reality. The dataset used in this paper can be found at https://github.com/canhld94/Daejeon520...

City-Scale Visual Place Recognition with Deep Local Features Based on Multi-Scale Ordered VLAD Pooling