Talks

How to deliver high-motion (sports) content to mobile device users and Smart TV in maximum quality. How to automatically transform low-motion content into high-motion content (frame doubling) in real time. How to implement end-to-end stream processing entirely in software using GPU acceleration.

The history of the creation of a service for regional FM radio stations. What were the tasks, problems and where it was possible to simplify life.

We will analyze the key problems of DPDK implementation. We will pay special attention to cases within SelectOS — a specialized OS where we implement DPDK-based functionality.

What is an IP camera, now widely used for video surveillance. How it is designed, starting with the chips, filter sensors and embedded processing algorithms. How the hardware works. Principles of sensor operation, nuances of operation and tuning algorithms for capturing and converting light flux. Interaction and control of optics, autofocus algorithms, backlighting, nuances of night and day shooting.

Development of Industrial Computer Vision Systems: using monocular depth estimation and object detectors, system optimisation and profiling

How to transcode video directly in the browser using WebCodecs, WebAssembly and other modern APIs to reduce server load and provide offline processing. I will analyse in which cases it really works effectively, what tools are already available in JavaScript and what limitations you will have to face in practice.

How do VK Calls work in poor network conditions? Let's break down the key techniques for stable audio calls under high packet loss and low bandwidth.

I'll tell you how we made our own transcoding with blackjack and adaptive at bench scale. I will touch on the topics of both hardware and software.

I will tell you how we solve the problem of quality assessment in the Unified Video Platform VK. I will touch on two main topics: calculating metrics for assessing the work of our transcoders and testing the player's work.

Comparison of video encoding quality on different hardware platforms. The Gisle Bjøntegaard method when comparing quality and its practical application.

We'll talk about video analytics using various hardware platforms to accelerate inference. We'll consider the deployment of models for different types of accelerators. We'll discuss tools for organizing video analytics pipelines. I will tell you about the development of our own video analytics pipeline and its modules.

I'll tell you how the short video feeds in Yandex (Market, Go, Rhythm) are designed "under the hood" and where and how to optimize their work.