说明：  presented dictionary pair classifierdriven CNNs for object detection, where dictionary pair back propagation (DPBP) is proposed for the end-to-end learning of dictionary pair classifiers and CNN representation, and sample weighting is adopted to improve the localization performance