TABLE 2. The Performance of Using Attention Transfer and Attention Transfer With Label Smoothing Regularizer (Attention Transfer+ LSR) for Classification of Blacklegged Ticks Versus Other Tick Specious. Teachers are Inception-Resnet Pre-Trained on ImageNet, and Students are Lighter CNN Model With Trainable Parameters. The Best Performances Per Each Column are in Bold. Smoothing the Loss Function Through LSR Approach Makes the CNN Model Perform Slightly Better on Accuracy Measure.

Model	Accuracy	ROC-AUC	PR-AUC
attention transfer	91.20 ± 0.33	97.70 ± 0.29	96.69 ± 0.08
attention transfer+ LSR	92.55 ± 0.39	97.32 ± 0.32	96.17 ± 0.05