Table 4 Comparison of the performance of different methods on the representing crowd counting datasets.
,
and
indicate the first, the second and the third best performance, respectively. Note that the MAE in WorldExpo’10 [55] is the average value of the five cross-scenes, SFCN\(\dag ^2\) [97] represents that model takes ResNet101 as backbone, pre-trained on GCC [97], “–” denotes that results are not available and “\(\downarrow\)” indicates the lower the better of the results [1, 5,6,7,8,9, 11,12,13, 16, 34, 38, 39, 41, 51, 52, 55,56,57,58, 84,85,86,87, 89, 92,93,94, 96,97,98,99,100,101,102,103, 105,106,107,108, 110,111,112, 159, 205, 220, 221, 228,229,230,231,232,233,234,235,236,237,238,239]
,
and
indicate the first, the second and the third best performance, respectively. Note that the MAE in WorldExpo’10 [From: A survey of deep learning methods for density estimation and crowd counting
