search

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

saving predicted keypoints into a Mat file. #40

Closed PastaAndBeer closed 2 years ago

PastaAndBeer commented 2 years ago

Hello. Thanks for sharing your great work.

I try to train SH on Human3.6M dataset. I wanted to see the predictions on imgs, so saved the preds which are used in mpii_eval function.

https://github.com/princeton-vl/pytorch_stacked_hourglass/blob/ed91059e874f35089dd3a8e692fa895929785b91/test.py#L230

The saved preds are actually so different from my thought. For example, array([[array([[ 2.57000000e+02, 6.73000000e+02, 2.11715072e-01], [ 2.52000000e+02, 2.44000000e+02, -0.00000000e+00], [ 2.57000000e+02, 6.73000000e+02, 1.27586707e-01], [ 2.57000000e+02, 6.73000000e+02, 1.27375469e-01], [ 2.52000000e+02, 2.44000000e+02, -0.00000000e+00], ...

May you please tell me how to get predicted joints coordinates and save them?

PastaAndBeer commented 2 years ago

Sorry, it seems like I did something wrong at training. I am going to see again.