Training method of crowd counting network

Abstract

The invention provides a training method of a crowd counting network, and the method provided by the invention effectively extracts multi-scale information through a plurality of groups of pyramid convolution kernels with different voidage rates, and solves the problem that the sizes of heads are not uniform. By adding batch normalization to each layer of output, the problem of difficulty in training caused by increase of the depth of the network is solved, meanwhile, the depth of the network is further improved through a residual structure under the condition that the parameter quantity is not increased, and high robustness is achieved.

Description

technical field [0001] The invention relates to a training method for a crowd counting network. Background technique [0002] Crowd counting and crowd localization is an important task in current computer vision. However, in actual situations, due to the variable shooting angles, the sizes of the heads in the pictures are not uniform, and there are serious occlusions and uneven distribution of crowds in high-density scenes, which will increase the difficulty of crowd counting and crowd positioning tasks. The emergence of convolutional neural networks provides a better way to achieve these two tasks. Usually we want the depth of the network to be as deep as possible to better map the relationship between input and output, but as the depth of the network increases, The amount of parameters will increase, making network training difficult, and even causing gradient explosion or gradient disappearance. Contents of the invention [0003] The object of the present invention is...

AI Technical Summary

Problems solved by technology

However, in actual situations, due to the variable shooting angles, the sizes of the heads in the pictures are not uniform, and there are serious occlusions and uneven distribution of crowds in high-density scenes, which will increase the difficulty of crowd counting and crowd positioning tasks.

The emergence of convolutional neural networks provides a better way to achieve these two tasks. Usually we want the depth of the network to be as deep as possible to better map the relationship between input and output, but as the depth of the network increases, The amount of parameters will increase, making network training difficult, and even causing gradient explosion or gradient disappearance

Images