Closed Rbrq03 closed 3 months ago
Hey there, congrats to the great work. Have you ever tried to train any version of RT-DETR from scratch? Will it drop some mAP comparing with training from a pre-trained backbone? I will be appreciate if you can share any train log.
We do not train model from scratch. I think it may need to tune data augmentation and optimizer and other superparameters carefully.
Thanks for your answer.
Hey there, congrats to the great work. Have you ever tried to train any version of RT-DETR from scratch? Will it drop some mAP comparing with training from a pre-trained backbone? I will be appreciate if you can share any train log.