Improvement of Pose Recognition by Sparse Regularized Convolutional Neural Network
Yuan Zhang, Yao Xiao, Zhongli Wang, Hao Su, Zihan Yu, Guanying Huo, Ning Xu, Xiaofeng Liu · 2019
The paper proposes a method of using deep model (CNN) with sparse regularization term to improve the performance of pose recognition. Convolutional neural network shows its limitations for pose recognition because of its bad convergence. It is believed that the activation function could be simplified by adding the sparseness term. Therefore, we present an algorithm applying the sparse regularization for the deep model with ReLU. Experimental results confirm that the proposed method can accelerate the convergence speed while a high recognition rate maintained.