mmsegmentation-distiller

mmsegmentation-distiller copied to clipboard

大佬，用cwd在目标检测上进行蒸馏的超参数是什么样。我用了detection仓库中配置文件给的t=1，α=5. 训练结果只有0.0几。

nameLiMing

Is this project still being maintained, please? I tried to join the wechat group chat via the QR code but never got through.

Geleven7

# Patching CVE-2007-4559 Hi, we are security researchers from the Advanced Research Center at [Trellix](https://www.trellix.com). We have began a campaign to patch a widespread bug named CVE-2007-4559. CVE-2007-4559 is a...

TrellixVulnTeam

cwd loss 和论文中的公式不一致

3

hi，请问作者 在论文里你用到了KL去学习Teacher的输出。 Equation (4)  在你的代码里 ` loss = torch.sum(-softmax_pred_T * logsoftmax(preds_S.view(-1, W * H) / self.tau)) * ( self.tau**2) ` 你使用的这个公式去计算KD loss。 但是标准的KL Loss应该是这样 ` kl_loss = torch.sum(softmax_pred_T...

Crazod

group wechat

4

Hello！ Recently, we are studying semantic segmentation and knowledge distillation. Can we update the QR code of the discussion group?

sunshiding

Wechat Group

1

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

pppppM

使用readme中的训练命令，loss不下降

13

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

GeneralJing

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

1

I have tested the pretrained Teacher model PSPNet+Res101 and got 73+ val mIOU. It is not consistent with the original paper. Could you help me? Thanks!

winycg

How to support pytorch1.9

1

Hi, I would like to ask how to support the pytorch1.9 in this method

Wangman1

how to distiller my own network structure?

1

when i use a different network, how to distiller my own network structure? can you supply some basic steps to do this? and i see the build_distiller function need teacher_cfg...

GeneralJing