search

leoxiaobin / deep-high-resolution-net.pytorch

The project is an official implementation of our CVPR2019 paper "Deep High-Resolution Representation Learning for Human Pose Estimation"
https://jingdongwang2017.github.io/Projects/HRNet/PoseEstimation.html
MIT License
4.31k stars 908 forks source link

关于论文的网络结构和实现的网络结构不同 #268

Closed zswldxb closed 2 years ago

zswldxb commented 2 years ago

1 中，从stage3 到stage4 的时候，如下图所示， 5。而实际的网络架构实现为 7

zswldxb commented 2 years ago

请问是这样的么，还是我理解错误了

unauna9739 commented 2 years ago

应该是这样的，然后stage3的模块化块重复4次，stage3到stage4的部分是transition层