. 2021 Dec 1;124:108473. doi: 10.1016/j.patcog.2021.108473

Table 1.

The achieved verification performance of different experimental settings by ResNet-100, ResNet-50 and MobileFaceNet models along with EUM trained with triplet loss and EUM trained with SRT loss. The result is reported on MFR dataset. The FMR100_Th^UMR-UMP are equal to 0.2307, 0.2652 and 0.3246 for ResNet-100, ResNet-50 and MobileFaceNet, respectively. The FMR1000_Th^UMR-UMP are equal to 0.3482, 0.3926 and 0.4476 for ResNet-100, ResNet-50 and MobileFaceNet, respectively. The lowest EER and the lowest average error of FMR100 and FMR1000 at the defined threshold for each of the evaluation cases are marked in bold. One can notice the significant improvement in the verification performance induced by our proposed approach (SRT) in most evaluation cases.

					FMR100_Th^UMR-UMP			FMR1000_Th^UMR-UMP
MFR	Setting	EER%	FMR100%	FMR1000%	FMR%	FNMR%	Avg.%	FMR%	FNMR%	Avg.%	G-mean	I-mean	FDR
	UMR-UMP	0.0000	0.0000	0.0000	1.0000	0.0000	0.5000	0.1000	0.0000	0.0500	0.8534	0.0252	70.7159
	UMR-MP	0.8914	0.8793	2.3347	0.4829	1.1886	0.8358	0.0082	6.0461	3.0272	0.5271	0.0203	15.0316
	UMR-MP(T)	1.0430	1.0794	4.7726	0.3084	2.4257	1.3670	0.0000	17.4773	8.7386	0.4331	0.0188	12.0587
	UMR-MP(SRT)	0.7702	0.6610	2.0558	0.4717	0.9460	0.7089	0.0108	4.8029	2.4068	0.5379	0.0221	15.9027
	MR-MP	0.8014	0.7695	1.3155	4.3230	0.5971	2.4601	0.4031	0.8685	0.6358	0.7314	0.0560	18.7469
	MR-MP(T)	0.9598	0.9471	2.6348	16.0855	0.4513	8.2684	2.4656	0.7660	1.6158	0.7415	0.1185	15.2544
ResNet-100	MR-MP(SRT)	0.8270	0.8015	1.4433	3.6616	0.6482	2.1549	0.3083	0.9994	0.6539	0.7248	0.0486	18.3184
	UMR-UMP	0.0000	0.0000	0.0000	1.0000	0.0000	0.5000	0.1000	0.0000	0.0500	0.8538	0.0349	55.9594
	UMR-MP	1.2492	1.4251	3.7780	0.4308	1.9709	1.2008	0.0007	10.6246	5.3126	0.5254	0.0251	12.6189
	UMR-MP(T)	1.9789	2.9533	7.9988	0.5626	4.0206	2.2916	0.0000	30.6549	15.3275	0.4401	0.0392	9.4412
	UMR-MP(SRT)	0.9611	0.9460	2.5652	0.5595	1.2129	0.8862	0.0030	7.4591	3.7310	0.5447	0.0272	13.4045
	MR-MP	1.2963	1.4145	2.6311	3.7683	0.8302	2.2993	0.2222	2.0467	1.1345	0.7232	0.0675	15.1356
	MR-MP(T)	1.3091	1.4560	2.8259	96.3681	0.0000	48.1840	62.1757	0.1980	31.1868	0.8269	0.4169	13.0528
ResNet-50	MR-MP(SRT)	1.1207	1.1367	2.4523	3.2837	0.8717	2.0777	0.2227	1.8775	1.0501	0.7189	0.0557	15.1666
	UMR-UMP	0.0000	0.0000	0.0000	1.0000	0.0000	0.5000	0.1000	0.0000	0.0500	0.8432	0.0488	37.3820
	UMR-MP	3.4939	6.5070	20.5640	0.2723	12.3833	6.3278	0.0088	40.4063	20.2075	0.4680	0.0307	7.1499
	UMR-MP(T)	5.2759	12.7835	28.8175	0.2151	21.7829	10.9990	0.0149	66.7192	33.3671	0.3991	0.0501	5.9623
	UMR-MP(SRT)	2.8805	4.6331	13.4384	0.3746	7.3802	3.8774	0.0097	30.1516	15.0807	0.5013	0.0383	8.6322
	MR-MP	3.5060	6.8842	17.3479	4.6039	2.8674	3.7357	0.5465	8.6723	4.6094	0.6769	0.1097	7.9614
	MR-MP(T)	4.2947	7.9124	16.3772	94.0982	0.0064	47.0523	61.3860	0.6354	31.0107	0.8082	0.4716	6.6455
MobileFaceNet	MR-MP(SRT)	3.1866	5.6166	13.5290	3.1906	3.1867	3.1886	0.2658	9.4802	4.8730	0.6636	0.0837	8.0905