pytorch_Realtime_Multi-Person_Pose_Estimation icon indicating copy to clipboard operation
pytorch_Realtime_Multi-Person_Pose_Estimation copied to clipboard

Published 20 hours ago • verified publisher icon tensorboy

Reproducing results

Open Vaufreyd opened this issue 7 years ago • 2 comments

Hello,

I thing I run the right way your code on my computer (with the last version of pytorch) but I did not get the same results. With VGG19, I stay stuck to mAP=0.51. I have few questions:

  • Did you run it as mentionned in the main page with batch_size=100? I did not manage to run it with more than batch_size=32 (with 2x 1080 GTX Ti in my computer). Do you have any idea of this problem?
  • How many epochs did you run?
  • Could you share the pytorch and other tool versions you used?

Thank you.

Vaufreyd avatar Sep 25 '18 06:09 Vaufreyd

I experience the same problem with the provided code and configuration. Val loss seems to be converging at epoch 60 where I only get 0.51 mAP. I train with: python train_VGG19.py --batch_size 100 --logdir {where to store tensorboardX logs}

Are you sure the training code train_VGG19.py is the one you used to obtain pose_model_scratch.pth ?

mihaifieraru avatar Feb 16 '19 16:02 mihaifieraru

same question

yedaorman avatar Apr 29 '19 01:04 yedaorman