An improved deep feature learning method for person re-identification
Jiazhen Xu, Chinyea Wang · 2017
Deep learning methods are widely used for person re-identification recently and outperform other methods. Models that combines identification and verification are proposed to learn feature embedding to satisfy intra-class compactness and inter-class dispension. However, sampling pairs in verification model remains complicated for it may not only create a large amount of redundant data but reduce the model accuracy. To this end, we propose an improved convolutional neural network whose verification model learns similarity by the distance between a training example and its center in favor of that of two examples so that no more sampling needed. Experimental results show that our method achieves state-of-the-art performance compared with other leading methods on the benchmark datasets such as Market-1501 and CUHK03.