Table 2.

Frame-level AUC performance comparison between two methods of prediction error encoding: PSNR and MSE. Results are reported on Avenue dataset using SVM classifiers. 80% samples of test set are used for training SVM.

CGAN Stream	MSE	PSNR
CGAN-1	0.72	0.82
CGAN-2	0.69	0.81
CGAN-3	0.73	0.79
CGAN-4	0.62	0.67
CGAN-(1 + 2 + 3 + 4)	0.81	0.92