Table 8 The MAE/RMSE of mainstream methods on the distractors of JHU-CROWD and the negative samples of NWPU-Crowd
From: A survey of deep learning methods for density estimation and crowd counting
Method | Distractors in JHU-CROWD | Negative samples in NWPU-Crowd |
|---|---|---|
MCNN [1] | 103.8/238.5 | 356.0/1232.5 |
CMTL [1] | 135.8/263.8 | - |
Switching CNN [5] | 100.5/235.5 | - |
SA-Net [11] | 71.9/167.7 | 432.0/974.4 |
PCC-Net [101] | - | 85.3/438.8 |
CSRNet [12] | 44.3/102.4 | 176.0/572.3 |
C3F-VGG [264] | - | 141.0/474.2 |
CANNet [99] | - | 82.6/343.4 |
SCAR [262] | - | 122.9/660.8 |
SFCN† [97] | - | 54.2/154.7 |
CG-DRCN [160] | 43.4/97.8 | - |