any progress on training code? about pytorch_realtime_multi-person_pose_estimation HOT 9 CLOSED

Hi, @mdering. Thanks for your attention for that reproduced code. I'm new for pose estimation and still studying how Part Affinity Fields works.

from pytorch_realtime_multi-person_pose_estimation.

that's too bad. the source material is somewhat ambiguous on their choices of thresholds and which space they're calculated in. Let me know if you find anything implemented in python

from pytorch_realtime_multi-person_pose_estimation.

Yes, I almost got crazy for their source code in here and the parameters are here (In caffe.proto file. message CPMTransformationParameter defined all the parameters for data augmentation).

Thank you @mdering !

from pytorch_realtime_multi-person_pose_estimation.

the code and parameters i get, but the augmentation stuff, and the masking procedures are very hard to parse in c++.

from pytorch_realtime_multi-person_pose_estimation.

I am also working on rewriting this code, for me the difficult part is the rtpose.cpp where the PAF calculation occurs, I get the big picture but many details remain to be clear . I think the cpm_transformation should be easier though might also takes some time.

from pytorch_realtime_multi-person_pose_estimation.

@ouceduxzk happy to know you also trying to reproduce it!

from pytorch_realtime_multi-person_pose_estimation.

Sounds great!

from pytorch_realtime_multi-person_pose_estimation.

an update to this, I still have not found a model that even comes close to converging during training. I am truly stumped.

from pytorch_realtime_multi-person_pose_estimation.

Try train.py

from pytorch_realtime_multi-person_pose_estimation.