I\'ve recently read blogs on RetinaNet and I have yet to understood the inner details of anchor target computation. So we have the multi-scale feature maps obtained from the FPN
