Our paper has been accepted for publication at the Conference on Computer Vision and Pattern Recognition (CVPR 2022).
Hyperbolic Vision Transformers: Combining Improvements in Metric Learning by Aleksandr Ermolov, Leyla Mirvakhabova, Valentin Khrulkov, Nicu Sebe, Ivan Oseledets
This study investigates how recent advances in computer vision can produce the most powerful metric learning model. By combining vision transformers and hyperbolic image embeddings, we achieved state-of-the-art results on four datasets. In some cases, we even surpassed the preceding metrics by a large margin. Our approach does not require any advanced training schemes and can be used in a plug-and-play matter in standard metric learning pipelines.