Detail of the student project

Topic:Efficient extraction of deep visual descriptors
Department:Katedra kybernetiky
Supervisor:Georgios Tolias, Ph.D.
Announce as:Diplomová práce, Bakalářská práce, Semestrální projekt
Description:Visual recognition of particular objects with deep learning is known to work better with high resolution images. Additionally, the deeper the model the better the performance. Both these aspects make the process of visual descriptor extraction to be computationally costly. The goal of this project is to mimic the result of very deep models and large resolution images, while using smaller models and small resolution images.
Bibliography:Radenovic Tolias Chum, PAMI 2019, Fine-tuning CNN Image Retrieval with No Human Annotation
Hinton, Vinyals, Dean, NeurIPS 2014 workshop, Distilling the Knowledge in a Neural Network
Responsible person: Petr Pošík