Usually the Teacher model performs higher results compared with the Student model. We used the Teacher model to report our results. Unbiased Teacher v1 also uses the Teacher model.
Also, the Teacher model might get zeros mAP during the burn-in stage, since it is randomly initialized without gradient updates or EMA updates from the Student.
Usually the Teacher model performs higher results compared with the Student model. We used the Teacher model to report our results. Unbiased Teacher v1 also uses the Teacher model.
Also, the Teacher model might get zeros mAP during the burn-in stage, since it is randomly initialized without gradient updates or EMA updates from the Student.