I trained the Ultralytics YOLOv5 model, https://github.com/ultralytics/yolov5/, on my custom dataset. Interestingly, when I run this trained model through the file val.py, the results I get do not seem like the bounding boxes I see as an output of detect.py.

I looked at the metrics in the table output by val.py at the end of its run and looked at some of the images that had bounding boxes drawn from the YOLOv5’s output.