Enabling Computer Vision Breakthroughs by Removing Computational Bottlenecks