Graduated from Lomonosov’s Moscow State University with a major in Physics. Has more than 20 years of software development background. Starting from 2011 specializes in applying machine learning and neural networks to solve typical problems in genetics, natural language processing and media processing to replace/improve known and widely used mathematical and statistical approaches.
Topic: Real-time video stream transformation with neural networks
Short Description: Interest to image transformation via neural networks surged recently after August 2015 publication of scientific paper of Gatys et al. where he and his colleagues proposed method of style transfer from one image to another. Number of papers were published on this topic afterwards and multiple implementations were developed by individual developers. In tooLook we aimed for performing such style transfer in real-time for video stream. Despite that there are multiple existing libraries for computer vision in general none of them proved to be suitable for the task. We will talk about our experience, problems we met and how we solved it.