. 2023 Dec 7;14:1201145. doi: 10.3389/fpsyg.2023.1201145

TABLE 2.

The mean classification accuracy of each AFERS on faces of the same expression.

Expression	Dataset	FaceReader			DeepFace
		N_images	Accuracy		N_images	Accuracy
Angry	Chen’s	19	0.895	0.647	19	0.737	0.618
	Tu’s	15	0.333	0.647	15	0.467	0.618
	KDEF	31	0.968	0.982	31	0.903	0.857
	RaFD	25	1.000	0.982	25	0.800	0.857
Happy	Chen’s	214	0.967	0.955	207	0.923	0.885
	Tu’s	54	0.907	0.955	54	0.741	0.885
	KDEF	59	1.000	1.000	59	1.000	1.000
	RaFD	39	1.000	1.000	39	1.000	1.000
Sad	Chen’s	34	0.706	0.587	34	0.441	0.478
	Tu’s	12	0.250	0.587	12	0.583	0.478
	KDEF	20	1.000	1.000	20	0.800	0.791
	RaFD	23	1.000	1.000	23	0.783	0.791
Surprised	Chen’s	78	0.923	0.932	76	0.724	0.652
	Tu’s	39	0.949	0.932	39	0.513	0.652
	KDEF	17	1.000	1.000	17	0.882	0.682
	RaFD	27	1.000	1.000	27	0.556	0.682
Total	Chen	345	0.928	0.890	336	0.818	0.765
	Tu	120	0.783	0.890	120	0.617	0.765
	KDEF	127	0.992	0.996	127	0.929	0.871
	RaFD	114	1.000	0.996	114	0.807	0.871

The mean accuracies on the two Eastern datasets (i.e., Chen’s and Tu’s) and on the two Western datasets (i.e., KDEF and RaFD) are also presented for comparison. With the agreement criterion of 0.9, DeepFace returned null results for 7 happy faces and 2 surprised faces in Chen’s dataset.