Obtaining YOLOv10 "box_oo" Loss During Inference or Validation Outside of Training

[unai-gurbindo]

Hi, I'm trying to obtain the YOLOv10 loss during inference or validation outside of training. Specifically, I'm interested in the "box_oo" loss. I've been looking into how v10DetectLoss and v8DetectionLoss work, but I haven't been able to make the necessary changes in BaseValidator to get the losses during validation. I also haven't been able to create a post-processing function for the results with the ground truth, as I don't fully understand what the inputs for v8DetectionLoss should be.

I've seen several open issues around this topic, and the only response seems to be that we need to interact with v8DetectionLoss, but has anyone managed to achieve this?