What strategies are used to optimize inference speed and memory usage in deployed computer vision models?