I understand that the YOLOv5 model predicts 25200 bounding boxes between all 3 levels of output. How many does the YOLOv6 model predict, if the input resolution is 640x640?
I don't know the capabilities of Yolo V6 in terms of bounding boxes.
Nevertheless, you can compare Yolo V5 and Yolo V6 in terms of FLOPs and parameter quantities.
According to this benchmark, Yolo V6 should be twice more powerful as Yolo V5 in terms of FLOPs and params, so you should have about 50000 bounding boxes but it is just a rough order of magnitude as it depends on many parameters (including images quality and quantity).