Multi-Scale Guided Attention Network for Crowd Counting

Table 9

Experimental results of different combinations of loss functions’ ablation.

Dataset	ShanghaiTech A		ShanghaiTech B
Evaluation	MAE	MSE	MAE	MSE
MSE Loss	85.0	150.4	11.6	15.8
SSIM + MSE Loss	70.3	147.8	10.5	14.4
MAE + MSE Loss	70.6	140.6	10.3	14.5
SSIM + MAE + MSE Loss	66.5	96.7	9.2	13.8