search

yyliu01 / PS-MT

[CVPR'22] Perturbed and Strict Mean Teachers for Semi-supervised Semantic Segmentation
https://arxiv.org/pdf/2111.12903.pdf
MIT License
187 stars 17 forks source link

about network #42

Closed ruizes closed 3 months ago

ruizes commented 4 months ago

Hello. Can I know why you choose the two teacher networks,Can I replace a teacher network with a network whose parameters are updated independently using SGD?

Thank you !

yyliu01 commented 3 months ago

Hi @ruizes

I apologise for the late reply, I was in a travel.

1). The two teachers based structure can improve the reliance of the pseudo label and further improve the generalisation when they meet the hard unlabelled samples.

2). Yes, you can.

Thanks for your interest.

Cheers, Yuyuan

ruizes commented 3 months ago

Thank you for your reply. Have a great trip!