This document summarizes a tutorial on large-scale visual recognition. The tutorial aims to provide tools for handling large datasets, including scalable image representations like VLAD and Fisher Vectors, and efficient matching and learning techniques. It also shows how large-scale retrieval and classification are converging, with retrieval becoming more machine learning-based and classification more cost-aware. Finally, it demonstrates that large-scale visual recognition does not require huge resources, with examples of searching 100 million images in 250ms and training the 2010 ImageNet challenge in days on a single server.