How to Use Computer Vision to Study Large Corpora of Images

This How-to Guide describes the use of computer vision techniques for analysing large corpora of images. Visual culture and its associated visual materials like photos, memes, emojis, diagrams, or maps are distinct features of online interaction. The guide introduces some of the available tools for identifying objects, moods, locations, and people in such material and discusses their respective strengths and weaknesses. Particularly, it distinguishes between commercial and open-source tools as well as between supervised and unsupervised machine learning approaches to image analysis. This leads to discussions about bias in training data, the possibility and feasibility of training your own models, and the scenarios in which models that do not require training data can be used. Finally, the guide provides some pointers on ethics and General Data Protection Regulation issues related to the use of computer vision in image analysis.