A DPM Based Approach to Joint Object Detection and Sub-category Recognition

Object detection and sub-category recognition play important roles in the field of computer vision. Most of the existing approaches separate detection and recognition into two sequential parts. We argue, however, detection and recognition could share information of each other to achieve a better performance for both of them. In this paper, a new approach to joint detection and recognition based on Deformable Part Model (DPM) is presented. Our approach extends DPM from pure object detection to simultaneous detection and sub-category recognition. A multi-objective optimization function is formulated. It integrates supervised sub-category recognition into DPM training process, using structural SVM with latent variables. The experiments show that our approach achieves a very exciting result in a challenging vehicle data set.