Multiple layer model for object detection and sketch representation
Wencheng Li, Xin Wu, Ling Cai, Fuqiao Hu, Yuming Zhao · 2016
In this paper we propose a multiple layer model for object detection and sketch representation. Unlike most traditional detection models focusing on the object localization, we investigate both the object detection and sketch representation within an unified framework. Based on the multiple layer architecture, our model can provide the sketch information of the detected object. Meanwhile, we generalize it from single scale structure to multiple scales, which efficiently saves time consumed in the image pyramids construction. To efficiently train the classifier at the top layer, we employ the stochastic gradient descent algorithm to minimize the training error and back propagate it to the bottom layer. The experimental results demonstrate that our model outperforms the conventional active basis model.