Canet: Context-Aware Loss for Descriptor Learning | IEEE Conference Publication | IEEE Xplore