Table 8.

Experiments with upper-body Re-ID and self-supervised pre-training (cross-domain Re-ID). Symbols: MSMT17 Merged—MSMT17, which combines training and test parts.

Algorithm	MSMT17 Merged → DukeMTMC-reID
	${\mathit{Rank}}_{\mathbf{1}}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$	$\mathit{mAP}\mathbf{(}\mathbf{\uparrow }\mathbf{)}\mathbf{,}\mathbf{\%}$
Baseline (upper-body)	50.97	33.57
+Res2Net50	53.34	34.73
+AugMix and JS Loss	57.12	37.69
+Self-supervised Pre-training	58.24	38.70

The best results by metrics are highlighted in bold.