yolov5 vs previous yolos

[readicculus]

Why is it so difficult to find any information anywhere that goes into detail or even just summarizes what approach yolov5 takes that builds off of the previous yolos? All I could find is someone saying there is not yolov5 paper. Even just a summary of developments to the actual detection train/inference additions would be great to add to the wiki. All the tooling integrations seem very useful and well documented in the wiki but I imagine I'm not the only one who wants to know what makes yolov5 different from v3 or 4(not just in terms of the results).