Information about the learning rate

[SebastianJanampa]

Hello,

Thank you for your work. I was reading the training hyperparameters, and i noticed that the learning rate for DINOv2 is greater than the learning rate for the decoder. In previous DETR models, the backbone usually has a smaller learning rate than the other parameters. Why did you choose a learning rate of 0.2 for dinov2.

rf-detr/rfdetr/config.py

Lines 55 to 57 in ed54776

	class TrainConfig(BaseModel):
	lr: float = 1e-4
	lr_encoder: float = 1.5e-4