Table 3 Summary of the recognition accuracies of ResNet18 models trained under varying ambient lighting conditions.

From: Accurate but fragile passive non-line-of-sight recognition

Model Without data augmentation With data augmentation
Test setting
0 1 2 3 0 1 2 3
ResNet18-0 0.8473 0.7184 0.7469 0.7758 0.8740 0.7152 0.7467 0.847
ResNet18-1 0.6998 0.8665 0.8766 0.8197 0.7208 0.8863 0.8955 0.8532
ResNet18-2 0.7045 0.8561 0.8747 0.8167 0.7149 0.8744 0.9050 0.8373
ResNet18-3 0.7806 0.7690 0.7997 0.8715 0.8322 0.8039 0.8292 0.8965
ResNet18-M 0.8608 0.8757 0.8894 0.8818 0.8828 0.8922 0.9065 0.8982
  1. ResNet18-0/1/2/3 is trained with 0, 1, 2, and 3 plates in the setting, respectively, and ResNet18-M is trained with all four measured datasets. The test setting is labeled by the number of plates used in the setup. Accuracies with/without data augmentation are listed in the left/right panel.