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
186 stars 17 forks source link

questions about Conf-CE loss #1

Closed Darcy103 closed 2 years ago

Darcy103 commented 2 years ago

I recently read your paper and it's a great job. I have a question about the Conf-CE loss in the paper. Can you explain what the c(w) formula means? I confused y~ and y^ in the c(w), formula(4), figure 2. question1 question2

yyliu01 commented 2 years ago

Hi,

Thanks for your interest. C(w) here denotes the pixel's confidence within the output above the positive threshold, which can be seen as "how teachers are confident for their produced pseudo-labels".

I apologize for the confusion, the y tilde denotes the teacher's (hard) pseudo-labels, and the y hat in here is the one-hot type prediction. In your first figure, \hat{y} is the one-hot output from the ensemble of teacher networks, while in the bottom figure, it denotes the student's one-hot prediction.

More details will be shown in the soon released code.

Darcy103 commented 2 years ago

I understand, thanks for your answer!

yyliu01 commented 2 years ago

you're welcome.