Table 2.

AP Scores across model types using the HRNet backbone

Model	W32-256	W32-384	W48-384
RGB	0.753 ± 0.102	0.762 ± 0.095	0.779 ± 0.093
Depth	0.778 ± 0.082	0.769 ± 0.090	0.765 ± 0.091
IR	0.739 ± 0.099	0.747 ± 0.095	0.758 ± 0.096
EIF-RGB-D	0.761 ± 0.086	0.765 ± 0.090	0.768 ± 0.093
EIF-RGB-IR	0.747 ± 0.092	0.765 ± 0.089	0.715 ± 0.121
EIF-D-IR	0.780 ± 0.087	0.760 ± 0.094	0.785 ± 0.083
EIF-RGB-D-IR	0.763 ± 0.088	0.753 ± 0.092	0.773 ± 0.083
IIF-1	0.785 ± 0.083	0.788 ± 0.083	0.793 ± 0.086
IIF-2	0.800 ± 0.076	0.784 ± 0.090	0.774 ± 0.094
IIF-3	0.792 ± 0.077	0.790 ± 0.080	0.776 ± 0.094
IIF-4	0.753 ± 0.099	0.761 ± 0.100	0.771 ± 0.097
LIF-1	0.769 ± 0.088	0.775 ± 0.091	0.798 ± 0.078
LIF-2	0.757 ± 0.101	0.775 ± 0.089	0.771 ± 0.095
LIF-3	0.775 ± 0.089	0.802 ± 0.077	0.811 ± 0.069
LIF-4	0.749 ± 0.099	0.765 ± 0.100	0.772 ± 0.089

Scores in bold are the best scores for each backbone.

E/I/LIF Early/Intermediate/Late Image Fusion.