YOLOv5: Listing All Potential Classes a Detection Could Be

[amybda]

In YOLOv5 I am running detection inference in a python notebook - and I can easily get the results in the tensor format or pandas format by using results.pandas().xyxy[0] or results.xyxy[0]. However, often the confidence for the class of each detection is not quite 100%, which implies that the model also calculates the confidence that a given detection is any other class. Is there a way to access this?

I know in YOLOv8 you can use r.probs to look at information such as the top 5 confidence classes etc., but even that only works on the classify function as opposed to detections. I'm unsure if this information is available in YOLOv5, or in YOLOv8 for the detect function. I would like to use this information to get the weighted class (weighted by confidence)! If anyone could help me with this matter I'd greatly appreciate it!