An improved deep multiscale crowd counting network with perspective awareness

Research on dense crowd counting method combining attention and hybrid dilated convolution

“…Figure 4 and Figure 5 respectively show the sample pictures on Part A and Part B datasets, their corresponding true value density maps and the predicted density maps generated by the network in this paper. [1] 110.2 173.2 26.4 41.3 MSCNN [3] 83.8 127.4 17.7 30.2 ResNeXtFP [4] 69.3 104.7 14.3 21.9 CSRNet [2] 68.2 115.0 10.6 16.0 ZHAI [5] 66.8 100.0 11.6 18.4 ZHUGE [6] 65. As can be seen from Table 1, compared with the excellent algorithms over the past years, the MAE value and MSE value have been improved to different degrees.…”