search

SangbumChoi / MobileHumanPose

This repo is official PyTorch implementation of MobileHumanPose: Toward real-time 3D human pose estimation in mobile devices(CVPRW 2021).
MIT License
235 stars 29 forks source link

When I run the training code according to your readme, a key error is reported. #5

Closed APeiZou closed 3 years ago

APeiZou commented 3 years ago

image When I run the training code according to your readme, a key error is reported. What’s wrong with this？

unoShin commented 3 years ago

Use LPSKI instead of LPSKINET

APeiZou commented 3 years ago

@unoShin when Use LPSKI，A new problem appeared, and the loss became more and more abnormal。 image

SangbumChoi commented 3 years ago

@APeiZou Please let me know the difference between epochs, not by batches. If this continues with further epochs that it must be serious bug.

unoShin commented 3 years ago

@SangbumChoi Actually it keeps happening until the end of the training. 1

SangbumChoi commented 3 years ago

@unoShin Thanks for reply. Let me check today and fix the bug, if you need this code urgently then please refer to the past commits.

unoShin commented 3 years ago

Also, if I run test.py using that model, the process is killed by out of memory. (It can be due to my hardware spec) 2 And if I change vis = True in test.py, then another problem appears. image

SangbumChoi commented 3 years ago

@unoShin @APeiZou Thanks for noting this error. All errors are fixed (since the error was just simply i returned intermediate value for debugging). Please enjoy the code and feel free to ask any bugs or questions.

https://github.com/SangbumChoi/MobileHumanPose/commit/fb16b4dec94f8b0e87d0a233606ed280c248939c

SangbumChoi commented 3 years ago

Also, if I run test.py using that model, the process is killed by out of memory. (It can be due to my hardware spec) 2 And if I change vis = True in test.py, then another problem appears. image

For the above issue I also think it is about hardware specs. And for the bottom issue please let me know if the error keep shows with the most recent branch

APeiZou commented 3 years ago

@SangbumChoi When I test the epoch 24，the error is big. image

SangbumChoi commented 3 years ago

@APeiZou Can you try --epochs 24-24 instead of just 24? and also next time please give me some additional information (such as training log). It would be more easy to find the errors.

wangm-word commented 3 years ago

hi,i I place the dataset as required,but when i train ,it has a bug,can you help me? image @APeiZou @unoShin

APeiZou commented 3 years ago

@wangm-word image have no Muco-3DHP datasets, modified as above picture.

wangm-word commented 3 years ago

thank you a lot ,but i have new proplem image

SangbumChoi commented 3 years ago

thank you a lot ,but i have new proplem image

this might because you didn't pull recent branch

SangbumChoi commented 3 years ago

Does all of you have fixed this issue? Can I close?

wangm-word commented 3 years ago

Does all of you have fixed this issue? Can I close?

I have saved this issue ,you can close it ,thank you vrey much!