Faster RCNN的检测蒸馏(分类、回归、Feature-level、Feature-level+Mask)
Thecodeisheavilyborrowedfrom:1.Distillationforfasterrcnninclassification,regression,featurelevelhttp://papers.nips.cc/paper/6676-learning-efficient-object-detection-models-with-knowledge-distillation.