Distillation-of-Faster-rcnn

Distillation for faster rcnn in classification,regression,feature level,feature level +mask

Detail in my csdn blog:

https://blog.csdn.net/qq_33547191/article/details/95014337

https://blog.csdn.net/qq_33547191/article/details/95049838

The code is heavily borrowed from :

1.Distillation for faster rcnn in classification,regression,feature level

http://papers.nips.cc/paper/6676-learning-efficient-object-detection-models-with-knowledge-distillation.pdf

2.Distillation for faster rcnn in feature level +mask

http://openaccess.thecvf.com/content_CVPR_2019/papers/Wang_Distilling_Object_Detectors_With_Fine-Grained_Feature_Imitation_CVPR_2019_paper.pdf

code:

https://github.com/twangnh/Distilling-Object-Detectors

main.py 里面介绍在哪里加入蒸馏

adaplayer.py 是针对teacher和student的feature map大小不相同时进行特征图大小进行变换

mimic_loss.py是如何计算分类和回归的蒸馏loss，其实就是计算他们的相似度

蒸馏在teacher和student的精度差距非常大时效果特别明显。