Transferable Feature Representation for Visible-to-Infrared Cross-Dataset Human Action Recognition

Table 1

The average recognition precision (%) of iDTs feature in XD145 and InfAR datasets. train_num and test_num denote the number of samples in training and testing sets, respectively.

Settings	Accuracies in XD145 (%)	Accuracies in InfAR (%)
train_num = 25, test_num = 20	90.58 ± 2.14	69.25 ± 2.71
train_num = 20, test_num = 20	90.00 ± 2.19	66.92 ± 1.37
train_num = 15, test_num = 20	86.83 ± 1.78	63.92 ± 2.24
train_num = 10, test_num = 20	83.58 ± 2.74	40.33 ± 7.88
train_num = 5, test_num = 20	77.25 ± 5.29	30.25 ± 5.41