Detail of the student project

Topic:Texture and shape bias of convolutional neural networks
Department:Katedra kybernetiky
Supervisor:Georgios Tolias, Ph.D.
Announce as:Diplomová práce, Bakalářská práce, Semestrální projekt
Description:Convolutional Neural Networks (CNNs) are commonly thought to recognise objects by learning increasingly complex representations of object shapes. Some recent studies suggest a more important role of image textures. This bias limits the performance of transfer learning to different tasks. The goal of this project is to better balance the bias between textures and shapes while training on the popular ImageNet dataset.
Bibliography:R. Geirhos, P. Rubisch, C. Michaelis, M. Bethge, F. Wichmann, W. Brendel, ICLR 2019, ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness

Radenovic Tolias Chum, ECCV2018, Deep Shape Matching
Responsible person: Petr Pošík