motivation to use ratio metric for anchor fitness calculation in autoanchor.py

[kimberlykang]

Before I read the "autoanchor.py" I thought that fitness calculation would be done with iou metric, because YOLOv2, 3, and 4 used IOU metric for fitness and also for the distance metric. I checked the code and found out that YOLOv5 use ratio metric and iou metric was commented. And for the k-means, YOLOv5 use scipy.cluster.vq.kmeans, which use Euclidean distance for distance metric.

I wonder what the motivation for YOLOv5 to use Euclidean distance for bbox k-means, and ratio metric for anchor fitness.