Making Computer Vision Computationally Efficient