A Semi-Supervised Object Detection Algorithm Based on Teacher-Student Models with Strong-Weak Heads

Semi-supervised object detection algorithms based on the self-training paradigm produce pseudo bounding boxes with unavoidable noise. We propose a semi-supervised algorithm teacher-student models strong-weak heads to cope this problem. The strong and weak of teacher model solve quality measurement problem label localization obtain higher-quality labels. student are decoupled reduce negative impact noise classification regression. reach 52.5 mAP (+1.8) PASCAL visual classes (PASCAL VOC) dataset even up 53.5 (+3.2) by using Microsoft common objects in context (MS-COCO) train2017 as additional unlabeled data. On MS-COCO dataset, our method also improves about 1.0 experimental configurations 10% COCO COCO-full labeled