A Stochastic Parallel Gradient Descent Algorithem for Pedestrian Re-identification

Abstract

Pedestrians re-identification is a hot topic in the field of computer vision. Convolutional neural network(CNN) has achieved a good recognition effect in pedestrian re-identification. However, CNN is computationally intensive because of the vast pedestrian data and the depth of CNN training. As the requirement of higher accuracy, the training always takes days and even week. In this paper, we proposed a parallel stochastic gradient descent(SGD) algorithm, where five-hierarchy parallel structure sets up blocks based on pedestrian attributes. Moreover, the interval for updating parameters is analyzed to optimize parameter selections. Momentum and adaptive adjustment of the learning rate are also adopted in training. Our experiments results show that this parallelization method successfully speeds up the training process by five times and surpasses state-of-the-art in accuracy as well.