|Description:||The goal of this project is to develop a system for visual recognition of specific pieces of artwork. This falls in the category of instance-level recognition, where one is dealing with a huge number of classes, eg 200,000 artworks in the MET museum. Recognition will be pursued through deep descriptors and knn classifiers. This is challenging task due to the large inter-class similarities and the long tail distribution, i.e. most classes has only a single training image. Deep descriptors will be learned in a metric learning fashion.