2019/08/18 by Måns Larsson, Larsson, Måns, Erik Stenborg +9 · 2 citations
Computer Science · Engineering · #68T45 #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.1908.06387
openalex publication_date 2019/08/18 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
Long-term visual localization is the problem of estimating the camera pose of\na given query image in a scene whose appearance changes over time. It is an\nimportant problem in practice, for example, encountered in autonomous driving.\nIn order to gain robustness to such changes, long-term localization approaches\noften use segmantic segmentations as an invariant scene representation, as the\nsemantic meaning of each scene part should not be affected by seasonal and\nother changes. However, these representations are typically not very\ndiscriminative due to the limited number of available classes. In this paper,\nwe propose a new neural network, the Fine-Grained Segmentation Network (FGSN),\nthat can be used to provide image segmentations with a larger number of labels\nand can be trained in a self-supervised fashion. In addition, we show how FGSNs\ncan be trained to output consistent labels across seasonal changes. We\ndemonstrate through extensive experiments that integrating the fine-grained\nsegmentations produced by our FGSNs into existing localization algorithms leads\nto substantial improvements in localization performance.\n