Table 2.

Experimental results of crowd anomaly detection algorithms using Deep Learning on UCSD datasets (Ped1 and Ped2).

	UCSD PED1					UCSD PED2
	FL AUC	FL EER	PL AUC	PL EER	TPR	FL AUC	FL EER	PL AUC	PL EER	TPR
[45]	92.10	16.00	67.20	40.10		90.80	17.00
[47]			59.00		53.00			61.00		81.00
[7]						90.80	17.10	$\mathbf{87.30}$	$\mathbf{19.40}$
[57]	91.40	15.60	69.10	39.30		91.10	16.10
[8]	93.20					92.10
[58]	92.60	11.20
[49]	94.30	10.00	70.30	34.00
[6]		19.00		$\mathbf{24.00}$
[60]	93.20	8.40			$\mathbf{83.00}$	93.90	7.50			$\mathbf{84.00}$
[62]	92.50	15.10	69.90	25.10
[64]	$\mathbf{98.40}$	$\mathbf{0.75}$				$\mathbf{98.50}$	$\mathbf{0.92}$
[65]	95.50	8.00	64.50	40.80		88.40	18.00
[66]	87.00	24.00	$\mathbf{85.00}$		81.30	88.00	24.40	86.00		81.90
[67]	96.80	7.00	70.80	34.00		95.50	11.00