When the model can't tell the dog from the ball

Ran Lyra's fetch session through MAKRR to detect the ball. It detected Lyra too. To be fair, she can look like a few other categories — furball, dust bunny, teddy bear. The model had options.

This is what happens when a model hasn't been trained properly: wrong categories, low confidence, confused detections. The fix is more data, better annotations, and better footage — enough that the system reliably tells a teacup Yorkie apart from a ball.