global_localization_mapper_node use cpu

[davidlmljll]

feature_extractor_ load image from cv_bridge,so feature_extractor_ is using cpu but descriptor_type and detector_type using SIFT_CV_CUDA_xxx/SUPER_POINT_xxx for GPU! So there is a conflict, how to solve it?

[willdzeng]

Hi @davidlmljll

feature_extractor_ interface does not support GPU image yet, it's taking a CPU image and convert it into GPU and send to SIFT_CV_CUDA internally, we might improve this later to avoid this extra copy.

Have you noticed significant performance degradation due to this copying?