2021/04/06 by Mehmet Özgür Türkoglu, Turkoglu, Mehmet Ozgur, Eric Brachmann +7 · 6 citations
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Advanced Vision and Imaging #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.2104.02538
openalex publication_date 2021/04/06 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Visual re-localization means using a single image as input to estimate the\ncamera's location and orientation relative to a pre-recorded environment. The\nhighest-scoring methods are "structure based," and need the query camera's\nintrinsics as an input to the model, with careful geometric optimization. When\nintrinsics are absent, methods vie for accuracy by making various other\nassumptions. This yields fairly good localization scores, but the models are\n"narrow" in some way, eg., requiring costly test-time computations, or depth\nsensors, or multiple query frames. In contrast, our proposed method makes few\nspecial assumptions, and is fairly lightweight in training and testing.\n Our pose regression network learns from only relative poses of training\nscenes. For inference, it builds a graph connecting the query image to training\ncounterparts and uses a graph neural network (GNN) with image representations\non nodes and image-pair representations on edges. By efficiently passing\nmessages between them, both representation types are refined to produce a\nconsistent camera pose estimate. We validate the effectiveness of our approach\non both standard indoor (7-Scenes) and outdoor (Cambridge Landmarks) camera\nre-localization benchmarks. Our relative pose regression method matches the\naccuracy of absolute pose regression networks, while retaining the\nrelative-pose models' test-time speed and ability to generalize to non-training\nscenes.\n