Date: Thursday December 7th
Time: 10:00 - 11:30 (GMT+1)
Since the seminal papers from Yann Le Cun, CNNs (convolutional neural networks) have been the workhorse behind deep learning based computer vision. They were responsible for the breakthroughs reaching better-than-human performance on fundamental computer vision tasks. Recently however, transformers have emerged as a powerful and flexible model. It first achieved breakthroughs in NLP but is now used in most cutting edge computer vision research. However, there still remain challenges in deploying them efficiently on resource constrained devices. In this webinar we discuss the state-of-the-art on how to design efficient vision transformer models, and how to deploy them on hardware. We conclude with some general observations on when it makes sense to use vision transformers today, and comment on future trends.