fotoscarreras.com contacted me because they had the idea to automatise the process of detect and recognise runners and cyclists in popular races. They shot thousands of photos in each event and to properly sell to each runner their photos they need to identify it to their race number. These was a manual process that takes days.
My proposition pipeline was to use face detection and create embeds with Facenet. Finally classify athletes creating face clusters. Despite helmets and sunglasses the results were very reasonable and around 90% of time was saved for each race.
Technology stack: python, MTCNN, pytorch Facenet
