If you were tasked with improving the accuracy of a computer vision model, what steps would you take?