Unstable training when using provided config file --> Early exiting due to MPJPE error

mkocabas / VIBE

Official implementation of CVPR2020 paper "VIBE: Video Inference for Human Body Pose and Shape Estimation"

https://arxiv.org/abs/1912.05656

Other

2.86k stars 550 forks source link

Unstable training when using provided config file --> Early exiting due to MPJPE error #208

Open grau4 opened 3 years ago

grau4 commented 3 years ago

Hi all,

I'm training VIBE using the provided config.yaml file (i.e datasets, hyperparams, seeds, etc). After the 9th epoch, I get the message:

"MPJPE error is 93.05937588214874, higher than 80.0. Exiting!"

which has been tagged as "error/pa-mpjpe" on the tensorboard writer. I wonder if anyone else has experienced this issue (?). Is there a trick on how to stabilize this loss? I'm not sure of why this loss/value has been used as exiting criteria.

Thanks! Best (I'm running on Windows)

AIML commented 2 years ago

Do you load pretrained regressor