Weights Standardization feature for better transfer learning.

[Shalom-P]

Hi I regularly train yolov5 detector models from scratch in a iterative manner for my job. Obviously we tried use the previous weights as the base weights and tried training the model from there but the model trained from scratch using the default weights performed better, but the drawback is that it takes a lot of time to train these models.
I recently came across weights standardization and how it better helps transfer learning I was wondering if you have support for such a function, obviously I am using a workaround for this but was just wondering if there was an existing solution for this problem.