search

Pose-Group / FAMI-Pose

This is an official implementation of our CVPR 2022 ORAL paper "Temporal Feature Alignment and Mutual Information Maximization for Video-Based Human Pose Estimation" .
46 stars 5 forks source link

Question About VSD #1

Open zhangrj91 opened 2 years ago

zhangrj91 commented 2 years ago

Thanks for your great work!

I have read your paper and I'm confused about the loss term of mutual formation. Specifically, I don't understand how to implement VSD into the term you have proposed when you take the label y as the condition. For example, I want to minimize the equation (7) so I will have to calculate the conditional distribution P(z|y)), but how can I get the latent representation z from the true label y?

zhangrj91 commented 2 years ago

I've understood.

qingshi9974 commented 2 years ago

I have the same question. Can you tell me how to calculate equation （7） through VSD?

zhangrj91 commented 2 years ago

你好，我是张仁杰，我已经收到你的邮件~我会尽快阅读你发来的邮件~

zhangrj91 commented 2 years ago

I have the same question. Can you tell me how to calculate equation （7） through VSD?

Sorry, I thought I have understood but I cannot reproduce, so I might be wrong maybe we should ask the author again.