Table 4.

Experimental evaluation of the algorithm with different backbones (cross-domain Re-ID). Symbols: D—DukeMTMC-reID; M—Market-1501; MSMT17 Merged—MSMT17, which combines training and test parts.

Algorithm	D → M		MSMT17 → M
	${\mathit{Rank}}_{\mathbf{1}}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	$\mathit{mAP}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	${\mathit{Rank}}_{\mathbf{1}}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	$\mathit{mAP}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$
Baseline (ResNet50)	54.30	25.50	58.82	30.25
With Res2Net50	54.16	25.95	62.02	32.70
Algorithm	M → D		MSMT17 → D
	${\mathit{Rank}}_{\mathbf{1}}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	$\mathit{mAP}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	${\mathit{Rank}}_{\mathbf{1}}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	$\mathit{mAP}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$
Baseline (ResNet50)	41.40	25.70	58.53	38.25
With Res2Net50	44.84	26.67	63.60	42.59
Algorithm	MSMT17 Merged → M		MSMT17 Merged → D
	${\mathit{Rank}}_{\mathbf{1}}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	$\mathbf{mAP}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	${\mathit{Rank}}_{\mathbf{1}}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	$\mathit{mAP}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$
Baseline (ResNet50)	65.65	37.71	66.16	47.74
With Res2Net50	69.74	41.32	70.65	52.37

The best results by metrics are highlighted in bold.