March 11-Deep High-Resolution Representation Learning for Human Pose Estimation-paper reading

https://blog.csdn.net/ls83776736/article/details/87993726
这篇文章的博主也是对于此文做了简单的介绍

论文原文下载

https://github.com/leoxiaobin/deep-high-resolution-net.pytorch
原文作者的代码在这里,这是CVPR2019年的大作~

下面这幅图是本文提出的HRNet 的模型的主要结构。

March 11-Deep High-Resolution Representation Learning for Human Pose Estimation-paper reading_第1张图片
这个模型和以前的模型不同,在模型中图像始终保持着高分辨率,但是平行的在subnetworks中,在向下的操作中,在做lower-resolusion,只是最后会讲这些与之前的network融合在一起。

下面对于网络依次介绍

March 11-Deep High-Resolution Representation Learning for Human Pose Estimation-paper reading_第2张图片
March 11-Deep High-Resolution Representation Learning for Human Pose Estimation-paper reading_第3张图片
在(2)可以看到,在最上面的那个分支中,图像的精度一直不变化,在second line, resolution,变为原来的一般,在third line,再减半,到fourth line, 再减半。

March 11-Deep High-Resolution Representation Learning for Human Pose Estimation-paper reading_第4张图片
March 11-Deep High-Resolution Representation Learning for Human Pose Estimation-paper reading_第5张图片
March 11-Deep High-Resolution Representation Learning for Human Pose Estimation-paper reading_第6张图片

This is the main architecture about HRNet, and more details should read paper again.

没有复现数据~

你可能感兴趣的:(Human,pose,estimation)