Learning Local Feature Descriptor with Motion Attribute for Vision-based Localization | Read Paper on Bytez