Normalizing Flow-Based Industrial Complex Background Anomaly Detection

Table 3

The AUROC score (%) on HSRBD datasets compared with the previous method.

Classes	GANomaly	SPADE	PaDiM	DifferNet	CS-Flow	CSA-Flow (ours)
Scenario 1	64.70	36.10	89.20	40.11	57.23	82.23 (−7.81%)
Scenario 2	68.70	86.00	96.00	69.56	98.67	100.00 (1.35%)
Scenario 3	76.00	77.40	96.60	73.56	94.23	100.00 (3.52%)
Scenario 4	57.00	94.50	67.30	60.61	83.03	89.09 (−5.72%)
Average	66.60	73.50	87.28	60.96	83.29	92.83 (6.37%)

Bold values denote the best result in the class.