CoPR: Toward Accurate Visual Localization With Continuous Place-Descriptor Regression

Visual Place Recognition (VPR) is an image-based localization method that estimates the camera location of a query image by retrieving most similar reference from map geo-tagged images. In this work, we look into two fundamental bottlenecks for its accuracy: sparseness and viewpoint invariance. Firstly, images VPR are only available at sparse poses in map, which enforces upper bound on maximum achievable accuracy through VPR. We therefore propose Continuous Place-descriptor Regression (CoPR) to densify improve accuracy. study various interpolation extrapolation models regress additional feature descriptors existing references. Secondly, compare different encoders show CoPR presents value all them. evaluate our three public datasets report average around 30% improvement VPR-based using CoPR, top 15% increase viewpoint-variant loss encoder. The complementary relation between Relative Pose Estimation also discussed.