R-SCoRe: Revisiting Scene Coordinate Regression for Robust Large-Scale Visual Localization

10 2 10 3 10 4 Map Size (MB) 0 20 40 60 80 Accuracy % Aachen Night (0.25m, 2°) Neumap HSCNet HSCNet++ ESAC (50) ACE (50) GLACE R-SCoRe HLoc+SPSG AS Cascaded QP+R.Sift Squeezer PixLoc Figure 1. Robust Visual Localization with R-SCoRe. Left: Point cloud of Aachen reconstructed by R-SCoRe. Right: On the large-scale Aachen Day-Night dataset [53, 56] using only daytime training images, R-SCoRe achieves 64.3% accuracy under the (0.25m, 2°) threshold for nighttime query images. It outperforms all previous SCR methods (circles) by a large margin. With a small map size of only 47MB at a comparable accuracy, R-SCoRe is an attractive alternative to traditional methods (triangles).