The model performance is evaluated in two aspects in this study: (1) accuracy, for which precision-recall curve, average precision (AP), mAP and missing rate curve are used, and (2) computation cost, for which detection speed and training time are used.