How to select the most suitable machine learning algorithm for a given computer vision problem?