CVPR2023
NeuMap: Neural Coordinate Mapping by Auto-Transdecoder for Camera Localization
Shitao Tang, Sicong Tang, Andrea Tagliasacchi, Ping Tan, Yasutaka Furukawa
摘要
Figure 1. The paper presents neural coordinate mapping (NeuMap) for camera localization. NeuMap encodes a scene into a set of codes and uses a scene-agnostic transformer to decode the coordinates of key-points in a query image. The right compares the localization accuracy and the data size for Aachen Night benchmark. The accuracy is averaged over three translation/rotation error thresholds (0.25m, 2 • ), (0.5m, 5 • ), or (5m, 10 • ). NeuMap and Squeezer control representation sizes and are illustrated by curves with multiple data points.