Transformers Neural Networks Applications in Different Computer Vision Tasks

Abstract

Transformers architectures are one of the latest inventions in the field of deep learning. Originally dedicated to NLP, they begin to find use in computer vision too. In this paper, we briefly describe the idea behind vision transformers and present a few examples, where we utilised them in our research, focusing on the field of medical images and autonomous driv-ing. We show, that vision transformers can be used in various tasks, such as detection or classification, as well as explain how some of their drawbacks can be mitigated with a transfer learning approach.