Crowd Counting Method Based On Improved Csrnet

Aiming at the problem of population counting, the research is getting deeper and deeper, CSRnet proposed a method of expanding convolution instead of convolution layer and pooling layer. This paper mainly proposes a multi-scale expansive convolutional neural network, which uses a method similar to the inception-ResNet module to calculate the population density of dense crowds with large scale changes and severe occlusion. And apply this method to ShanghaiTech dataset. The experimental results show that compared with CSRnet, the accuracy of this method has been improved, and the speed of feature extraction has also been improved.