microsoft / MeshTransformer

Research code for CVPR 2021 paper "End-to-End Human Pose and Mesh Reconstruction with Transformers"
https://arxiv.org/abs/2012.09760
MIT License
614 stars 95 forks source link

Question about the learning rate adjustment strategy #86

Closed delaprada closed 12 months ago

delaprada commented 12 months ago

Hi, thank you for providing this good work!

I have a question about the adjust_learning_rate method in run_metro_handmesh.py. I use metro to train from scratch with Freihand dataset. I found that the learning rate doesn't change and the value is always 0.0001. I refer to the freihand_reproduce_log.txt and also find that the learning rate is 0.0001 for the first 100 epochs and 0.00001 for the latter 100 epochs. It seems the adjust_learning_rate method in run_metro_handmesh.py doesn't work?

Thank you for your time and help!