search

zzh8829 / yolov3-tf2

YoloV3 Implemented in Tensorflow 2.0
MIT License
2.51k stars 913 forks source link

Patience conflict on early stopping and learning rate reduce #181

Open jrrios opened 4 years ago

jrrios commented 4 years ago

https://github.com/zzh8829/yolov3-tf2/blob/master/train.py#L177

ReduceLROnPlateau has a default patience of 10 epochs and the patience for EarlyStopping is set to 3.

This means the learning rate will not be reduced if the loss plateaus because it will early stop before it has a chance to update the learning rate

JimEverest commented 4 years ago

I found this issue too, so I set the patience for EarlyStopping to 6 and patience for ReduceLROnPlateau to 3.

jrrios commented 4 years ago

IMO it should be configurable for the user