How did we classify the 48 million roofs in France by their orientation and shape with minimum effort and time? First, we designed a gamified crowdsourcing platform to manually classify about 10,000 roofs. Then we used standard machine learning techniques (logistic regression and deep learning) to train an automated classifier. Applications include potential estimation of vegetalization and solar energy. The approach to augmenting human contribution impact with image processing techniques could be applied to a wide range of problems.