Neutral networks for computer vision

Computer vision is one of the fastest-growing fields in engineering. It allows to automatically analyze and understand images
and videos, which take up a significant share of about 80% of worldwide Internet traffic.

This course continues directly from the “Introduction to computer vision” course, and focuses, naturally, on neutral network-based methods for computer vision uses. The course is meant for students in all software study tracks with an appropriate background, specifically the machine learning track. It also addresses contemporary technologies at the forefront of industry and research.

The course will follow a similar structure to the introductory course, but one less intensive in terms of number of assignment – while equally in-depth and advanced in terms of professional content.

The main course subjects include: Reviewing and expanding ondetection, segmentation, temporary networks (recurrent neural networks, RNN), attention, transformers, image generation networks (GAN), 3D vision and video analysis networks, and reinforcement learning.

The course includes a weekly frontal lecture (2 academic hours), and a weekly lab/practice session (2 academic hours).

In the lab, students will perform and experience practical aspects of deep learning-based computer vision, while being exposed to scientific articles and sources on one hand, and to coding and practical considerations on the other. The computer exercises
will be done in a Python environment with Jupyter (local or on Google Colab) while using the appropriate libraries.