Missing annotations: towards automated gesture recognition on artworks

In the context of Computer Vision applied to Digital humanities and Art history, we currently miss a proper training dataset of artworks for body/object recognition that addresses the diversity and uncertainty linked to the large spectrum of non-naturalistic artistic depictions.
In this talk, I introduce and discuss the possibility of using Data Augmentation on an existing labeled dataset of real images in order to generate look-alike paintings as an alternative path. Finally, I introduce a new project called PAPA (Pose Annotations Project for Artworks), which goal is to produce an annotated dataset for body pose recognition on artworks via the help of a participatory platform. I herewith aim to engage the conversation on the creation of such a dataset, if it should require a specific ontology and what could be the benefits for this field of research.