search

open-mmlab / mmskeleton

A OpenMMLAB toolbox for human pose estimation, skeleton-based action recognition, and action synthesis.
Apache License 2.0
2.89k stars 1.02k forks source link

Segmentation Fault (core dumped)when testing pretrained models #87

Closed jfgg95 closed 5 years ago

jfgg95 commented 5 years ago

Code version (Git Hash) and PyTorch version

PyThorch 0.4.1

Dataset used

Kinetics-skeleton

Expected behavior

To evaluate ST-GCN model pretrained on Kinetics-skeleton

Actual behavior

strace python main.py recognition -c config/st_gcn/kinetics-skeleton/test.yaml

20180910222733

Can someone please help ?

jfgg95 commented 5 years ago

I have found the reason，this problem is caused by mkl.

yuan-2703 commented 4 years ago

I have found the reason，this problem is caused by mkl.

i met the same problem after running mmskl by using pytorch 1.3.1,Kinetics-skeleton Can you share your solution?

jfgg95 commented 4 years ago

i met the same problem after running mmskl by using pytorch 1.3.1,Kinetics-skeleton Can you share your solution?

This problem is caused by an old version of mkl installed by default in conda. I updated the mkl to the new version and solved the problem.