Schedule

Over the last couple of years, diffusion models have made tremendous progress in the field of video generation: from the first model trained on toy datasets to Sora. In this talk, we will analyse their journey and discuss the future prospects of the field.

I will talk about the NISQA-s neural network metric for assessing streaming voice quality. We will discuss how it differs from the original NISQA metric, areas of use, the benefits it can bring, and potential methods of application. It will be useful for developers of conferencing applications and other audio streaming systems.

We will talk about JPEG AI — the first international standard for image compression based entirely on neural networks. We will discuss a distinctive feature of the new standard — multitasking, which allows us to successfully solve not only image compression tasks, but also computer vision and quality improvement ones. I will tell you how the development took place and what you can expect from the implementation.

The magic of AI in cinema. I will share my experience of developing a complex system that turns actors' speech into accurate and synchronised subtitles using machine learning.

We will discuss the present and future of multimodal architectures based on language models in the task of describing videos and answering questions about them.

I will tell you about a system for identifying people by silhouettes that works even with partial or complete concealment of the face. The technology uses computer vision and machine learning to recognise silhouettes as well as to collect anonymised attributes.

We will look at different ways of estimating grain size distribution and discuss the advantages of using machine vision. I will talk about the architecture of the solution and the models used, and share tricks for separating granules on the photo for correct processing.

Standards, JPEG AI, and attacks.

How to increase viewing and realise that ML alone is not enough. Business rules, ML, storefronts, Python in enterprise.

We will consider the development of AI technologies in retrospect. We will focus on multimodal AI systems and intelligent agents that integrate information from different modalities. In conclusion, we will discuss the impact of advanced technologies on today's reality, as well as their future potential and the risks that inevitably arise.

Technical details of extracting features from video files to organise searches across a video archive of 20,000 videos.

In my talk, I will describe how we designed the vision system for our new unmanned platform. What devices we developed, what tools we used for modeling, and what final result we got.

Real cases of machine vision implementation at mining enterprises and Russian plants. Features of equipment selection and training of neural networks for work in difficult conditions.

How to use multiple ML models to automate the verticalisation of short videos for their viewer service.

On finding a real-time noise cancellation solution and what research we did to get a cleaned speech signal with improved speech quality.

Have you ever wondered how the magic works that changes the resolution of video and ruins your screen sharing?

How we at Mayflower built our Live Streaming Platform Engine from a DevOps perspective.

The talk is dedicated to real-time streaming with adaptive bitrate from video conferencing servers to users. It examines methods for estimating client network bandwidth on the server side, including the mathematical models and algorithms used. Particular attention is given to the architecture for integrating this solution into the product, which helps improve video call quality under fluctuating network conditions.

I'll talk about the bottlenecks of video conferencing applications in terms of CPU and GPU efficiency and the technologies we're abandoning the Qt framework in favour of.

Telemost lived on the open source media server Jitsi. But the time came when we had to write our own solution and run it for hundreds of millions of users. In the talk, you will learn why and how we did it. 

We will tell you about interesting and difficult moments in the development of P2P solution for traffic delivery on the streaming platform goodgame.ru.

Using the features of VK Calls as an example, I will explain how to bypass the limitations of WebRTC, Android SDK, and OpenGL.

I will talk about the NISQA-s neural network metric for assessing streaming voice quality. We will discuss how it differs from the original NISQA metric, areas of use, the benefits it can bring, and potential methods of application. It will be useful for developers of conferencing applications and other audio streaming systems.

I'll talk about the bottlenecks of video conferencing applications in terms of CPU and GPU efficiency and the technologies we're abandoning the Qt framework in favour of.

In Yandex we've developed our own means and technologies for quality control of smart devices based on Android including QC-App application. The talk is dedicated to the manufacturing process of our smart devices and to the contribution of Android developers to this process.

How we at KION created own Remote Config and partitioner for A/B experiments (analogue to Firebase).

A story about video archive for iGaming business. Why sometimes it is not enough to record video of what is happening and what else you need to save if you want to retrospectively analyse what happened in the future.

SDI has for many years been the only way to transmit uncompressed digital video. Despite this, it has fallen into the trap of being expensive: the rarity of use means low quality hardware. SMPTE 2110 was supposed to replace the rare hardware with publicly available reliable Ethernet. In the talk, I will tell you what its fatal irrecoverable flaws are.

The talk explores the evolution and current state of URLLC systems, highlighting how open source projects can be used to create cost-effective and flexible alternatives to commercial systems like DJI and Walksnail for reliable low-latency video transmission. It also addresses the engineering challenges and solutions in this field.

I will tell you how we reduced the number of video freezes among VK Video users. How we chose and fine-tuned metrics to evaluate the success of our experiments and what results we achieved.

A story about video archive for iGaming business. Why sometimes it is not enough to record video of what is happening and what else you need to save if you want to retrospectively analyse what happened in the future.

Out of the talk you will get the basic understanding of separate and joint source-channel coding principles, how it can help ultra-low latency video communication, and prospects of technology development in this direction.

SDI has for many years been the only way to transmit uncompressed digital video. Despite this, it has fallen into the trap of being expensive: the rarity of use means low quality hardware. SMPTE 2110 was supposed to replace the rare hardware with publicly available reliable Ethernet. In the talk, I will tell you what its fatal irrecoverable flaws are.

I will tell you how we reduced the number of video freezes among VK Video users. How we chose and fine-tuned metrics to evaluate the success of our experiments and what results we achieved.

We will tell you about interesting and difficult moments in the development of P2P solution for traffic delivery on the streaming platform goodgame.ru.

How the digital TV signal spreading all over the country is organised. Constant bitrate, metadata packaging next to video streams, extremely strict formalised requirements for each bit, myths about PCR — this is what our story of federal TV is about.

We will tell you how to make video playback fault-tolerant (survive connection breaks, 4xx and 5xx errors) and teach the player to switch between CDN nodes to redistribute the load.

We will consider the modern Content Steering specification for HLS and DASH standards, its convenience, applicability and alternatives using the example of Yandex video platform.

Standards, JPEG AI, and attacks.

There is a problem: how to render a user interface into an H.264 video stream. We will see how to do it efficiently and review details of RTP/SAVP protocol (used for media in real-time communication solutions such as Yandex Telemost) and H.264 format that help us achieve it.

We will tell you about non-obvious problems we faced during transcoder development, as well as about peculiarities of operation and optimisations of the implemented player.

Out of the talk you will get the basic understanding of separate and joint source-channel coding principles, how it can help ultra-low latency video communication, and prospects of technology development in this direction.

In our video player development project, we reduced time to market by implementing effective testing patterns. A key element of our strategy was automation using the Frankenstein framework and Kotlin Multiplatform, as well as the introduction of Continuous Integration (CI). These measures allowed us to deliver the product to users faster and improve the management of development processes.

We will tell you about non-obvious problems we faced during transcoder development, as well as about peculiarities of operation and optimisations of the implemented player.

We will tell you how to make video playback fault-tolerant (survive connection breaks, 4xx and 5xx errors) and teach the player to switch between CDN nodes to redistribute the load.

We will consider the modern Content Steering specification for HLS and DASH standards, its convenience, applicability and alternatives using the example of Yandex video platform.

How television began, how filming and editing took place, what were the first graphic stations and how today all this equipment can get a second life.

Review lecture on modern astronomical methods and technologies processing of optical observations and obtained digital data.

TBD

Our experience of creating a system for collecting and processing product events in the KION online cinema. Details about processing, validation and data quality.

How we at KION created own Remote Config and partitioner for A/B experiments (analogue to Firebase).

I will tell you about automation of video streaming testing. What tests have already been automated, what difficulties there were on the way and how we tried to solve them.

In our video player development project, we reduced time to market by implementing effective testing patterns. A key element of our strategy was automation using the Frankenstein framework and Kotlin Multiplatform, as well as the introduction of Continuous Integration (CI). These measures allowed us to deliver the product to users faster and improve the management of development processes.

Are you going to create a video editor using web technologies?

Sooner or later you will face the task of generating MP4. I'll tell you how and why we at Lumen5 redid our rendering several times and came back almost exactly where we started.

We will be talking about the schedule, sessions, and activities. Connect to the broadcast to find out what's in store for you!

We will be summarising the results, recall the highlights and talk about future plans. Connect to the broadcast to make sure you don't miss a thing!

We will be talking about the schedule, sessions, and share information. Join us in the hall or online to find out what's in store for you!

We will be summarising the results of the conference, recall the highlights and talk about future plans. Join us in the hall or online so you don't miss a thing!

We will discuss testing and debugging of video streaming in Smart TV applications. Actual tools and approaches for platforms with high fragmentation and complexity of automation.

We will discuss how different video services cope with high load, taking into account their specifics. We'll talk about traffic, hardware and preparing the team for incidents.

Lightning talks are a great format to discuss a topic in a dynamic way and find like-minded people. Fifteen-minute presentations on professional or near-professional topics and lively discussions are waiting for you. Have you long wanted to try your hand at being a speaker? Are you eager to share ideas or cases from your working practice with the world? Have you watched VideoTech presentations and feel that you can add something of your own to the reports? Then apply in advance or sign up to speak at the venue!