[help wanted] console log freezes while train... about mmpose HOT 5 CLOSED

Please try using workers_per_gpu=0 in the config file.

from mmpose.

We need more info to reproduce this.

Some simple checks before that:

• data, could you double check if your data is well prepared? you may try to truncate the person_keypoints_val2017.json to a few samples only, and run on this tiny subset.
• cpu/gpu status, check htop for cpu usage, and watch nvidia-smi for gpu usage. anything took 100% usage? How many gpus are used? is it runned in a docker env?
• 1h is surely abnormal. if nothing happens after 5min when debugging, you may interrupt it

ps. i edited the above post to fence the codes.

from mmpose.

We need more info to reproduce this.

Some simple checks before that:

• data, could you double check if your data is well prepared? you may try to truncate the person_keypoints_val2017.json to a few samples only, and run on this tiny subset.
• cpu/gpu status, check htop for cpu usage, and watch nvidia-smi for gpu usage. anything took 100% usage? How many gpus are used? is it runned in a docker env?
• 1h is surely abnormal. if nothing happens after 5min when debugging, you may interrupt it

ps. i edited the above post to fence the codes.

many thanks for your reply!

i print the log, find that it get stuck in epoch_based_runner.py(line 30),
when it fetch image
i have checked mine images ,but they are normal......

from mmpose.

Please try using workers_per_gpu=0 in the config file.

it works!!!!!👍👍👍👍👍 many many many thanks!!!!
i am so stupid~~~

from mmpose.

Please open a new issue instead of bump a closed one.
Their underlying cause might be different.

from mmpose.