Requesting more training details

[slimwangyue]

Thanks for your brilliant work. Could you provide more training details? In your paper you mentioned that to fully train a model, you first train from scratch then use data augmentation whose details are described in appendix. However, I did not find it in your arxiv paper. I am currently trying to apply knowledge distillation to your work. Is there any insights you could provide?

[WongKinYiu]

Hello,

you could find the training details in supp (https://openaccess.thecvf.com/content/CVPR2021/html/Wang_Scaled-YOLOv4_Scaling_Cross_Stage_Partial_Network_CVPR_2021_paper.html).

[slimwangyue]

@WongKinYiu I saw it. But it does not include the initial learning rate info.

[WongKinYiu]

https://github.com/WongKinYiu/ScaledYOLOv4/blob/yolov4-large/data/hyp.scratch.yaml#L6