pppppM

Results 84 comments of pppppM

Is the Experimental settings same as the paper？

Sorry, it is because the released edition wasn’t the code used in our paper. But later we would reproduce the results in our paper with the released code. And we...

cwd loss 和论文中的公式不一致

如果把 kl 的公式展开，和 student 无关的项省掉后（teacher 部分不影响梯度更新，只影响数值），就是代码里的形式

Why pretrained Teacher model PSPNet+Res101 cannot the reach the 78mIOU?

Thanks for your attention ! If you want to distill model in OpenMMLab related repos, could join the wechat group in README.md

group wechat

Updated. @sunshiding @Chan-Sun

group wechat

Updated @Nireil

ChannelWiseDivergence unused variable

Thanks for your attention ! The arg is deprecated. If you want to distill model in OpenMMLab related repos, could join the wechat group in README.md

使用readme中的训练命令，loss不下降

> 使用train中的代码进行蒸馏训练，看日志loss开始会下降一点，后面一致在44.几震荡，感觉不收敛，是设置的不对吗 loss 数值大是正常的，因为在实现的过程中，kl 计算时把没有梯度的项给去掉了，所以会导致数值有些大

使用readme中的训练命令，loss不下降

Thanks for your attention ! If you want to distill model in OpenMMLab related repos, could join the wechat group in README.md

How to support pytorch1.9

Thanks for your attention ! If you want to distill model in OpenMMLab related repos, could join the wechat group in README.md

how to distiller my own network structure?

Thanks for your attention ! If you want to distill model in OpenMMLab related repos, could join the wechat group in README.md