Specifying backbone learning rate for retraining DETR

Hi!

I am trying to retrain DETR for object detection. The DETR paper used a separate learning rate for the transformer and the backbone.

When I specify the learning rate in the training arguments, which learning rate is this in actuality?
How can specify multiple specific learning rates?

Thank you in advance!

1 Like

Hi I know this is very old, but did you ever figure this out? I am trying to do the same thing with RT-DETR.

1 Like

For now, I’ve looked for information.