Monday, September 16, 18:30-20:00
Ievgen Gorovyi (PhD, CEO @ It-Jim) and Pavlo Vyplavin (CTO @ It-Jim)
Information extraction from images is a rapidly growing research field. Nowadays, deep learning (DL) methods are demonstrating remarkable results in many computer vision (CV) problems like object detection, face and text recognition, action recognition, object recognition and object tracking. However, such methods require high amount of training data and a lot of computational resources like GPUs or even clusters of hardware units. Moreover, another drawback is lack of scene geometry information utilized in DL models, which may degrade the expected results. In contrast, classical methods are based on feature crafting and a batch of user defined parameters, which often leads to unstable results. In this tutorial, we will discuss several typical CV problems and their classical and DL solutions. Comparative analysis will allow to give a clear picture of what group of methods should be applied and when. In particular, we are going to cover problems of object detection, multi-view matching, object tracking and simultaneous localization and mapping (SLAM).
Ievgen Gorovyi received his PhD in 2014. His thesis was devoted to image correction and autofocusing methods for SAR systems. After receiving his PhD, he established It-Jim company, which delivers technical solutions in computer vision and pattern recognition, signal and image processing fields for clients over the globe. Ievgen is co-author of more than 40 publications, winner of numerous international awards and speaker at both academic and
Pavlo Vyplavin received his PhD in 2011 with a thesis devoted to ground based noise waveform SAR. He worked at IRE NASU, Ukraine and at University of Campinas, Brazil. In 2018 he joined It-Jim focusing his work on computer vision, signal processing and artificial intelligence. Pavlo is a co-author of more than 60 published works.