Improving the performance of multimedia pedestrian classification with images synthesized using a deep convolutional generative adversarial network

Abstract

Generative adversarial networks (GANs) are used to improve pedestrian identification performance and reduce labelling work, which is a significant burden on deep learning applications. As the input data are images, a deep convolutional generative adversarial network (DCGAN) is used, as the convolutional operation performs better than a typical GAN. The results produced using the proposed method are evaluated against the performance of a convolutional neural network according to the ratio of real and synthetic images generated as a result of training. Classification performance as a result of this ratio is analyzed; this will aid many developers who may use the results of this paper to train systems efficiently with less data.