EXPLOITING EFFECTS OF PARTS IN FINE-GRAINED CATEGORIZATION OF VEHICLES

Fine-grained categorization has become a hot topic in computer vision. Based on the theory that part information is crucial for fine-grained categorization, we proposed a part-based categorization method for vehicles, consisting vehicle parts localization, part-based vehicle representation and classification. There were three contributions we made in this work: 1) we analyzed discriminative powers of parts for fine-grained categorization; 2) we proposed a frame of how to integrate discriminative powers of parts into categorization, and proved that it can achieve better performance than treating every part equally; 3) we provided an annotated dataset with parts for vehicle categorization.