mmpose icon indicating copy to clipboard operation
mmpose copied to clipboard

Published 20 hours ago • verified publisher icon open-mmlab

Loading checkpoint is too slow

Open bzfather opened this issue 3 years ago • 3 comments
trafficstars

I firstly trained the model on servers and now I try to load it on desktop (Windows with 3090 CUDA11.4 pytorch1.10) I ran the demo and it was stuck at “Loading checkpoints.....” shown in picture. It took 20 minutes to load. What happened? 20220401_154439

bzfather avatar Apr 01 '22 13:04 bzfather

Thanks for your feedback. Does it always happen, or randomly happen? I did notice that in Windows sometimes the program gets stuck and needs to click Enter to continue, and it happens not only to mmpose but also other programs like git. Not sure if it is related to your issue.

ly015 avatar Apr 02 '22 03:04 ly015

Thanks for your feedback. Does it always happen, or randomly happen? I did notice that in Windows sometimes the program gets stuck and needs to click Enter to continue, and it happens not only to mmpose but also other programs like git. Not sure if it is related to your issue.

It always happened. It was stuck in this step when loading other checkpoints file as well. No reaction when clicking Enter. And when I checked task manager, I found that the memory of GPU increasing very slowly and only 2 cores of CPU working. After a long time of loading, it finished.

bzfather avatar Apr 03 '22 21:04 bzfather

It could be an MMCV issue. Could you please provide the environmental information by python -c "from mmcv.utils import collect_env; print(collect_env())"?

ly015 avatar Apr 05 '22 06:04 ly015