为了探究不同的深度卷积神经网络在行人检测任务中的性能差异,基于Faster-R-CNN深度学习算法框架,在Caltech行人数据集上对VGG-Net(Visual Geometry Group Net)和Res-Net(Residual Net)的性能进行了比较.通过改变数据集、改变训练数据的数量、对比训练过程中各阶段的检测率,对两个网络的泛化能力、学习能力以及收敛速度进行了对比.实验结果表明,Res-Net相比于VGG-Net网络具有更快的收敛速度和更强的泛化能力;Res-Net的学习能力更强,随着训练数据的扩展,其性能提升更大.在行人检测任务中,Res-Net具有更好的性能.%In order to explore the performance of different convolutional neural networks in the pedestrian detection task,VGG-Net and Res-Net based on the Faster-R-CNN framework with the Caltech database are implemented.By constructing extensive experiments through altering the database and changing the quantity of the train data and comparing the detection rate of each training stage,the generalization ability,learning capability and convergence rate of the two deep architectures are compared.Experimental results show that the Res-Net has faster convergence speed and stronger generalization ability than VGG-Net,and the Res-Net has stronger learning ability,it outperforms VGG-Net when the data scale increases.
展开▼
