princeton-vl / pytorch_stacked_hourglass

Pytorch implementation of the ECCV 2016 paper "Stacked Hourglass Networks for Human Pose Estimation"
BSD 3-Clause "New" or "Revised" License
465 stars 94 forks source link

2D joints as outputs #35

Closed dvdblk closed 3 years ago

dvdblk commented 3 years ago

Hello. If I understood correctly, this network outputs a set of 64x64 heatmaps for each joint. However, I would like to get 2D joint keypoints as the output.

In other words the output size I want would be (16, 2) = 16 times 2D coordinates of each joint.

What should I do in order to get this output from the heatmaps / network? :)

Thanks!!

crockwell commented 3 years ago

You are correct. Actually keypoints are generated via post-processing -- see test.py for details