Hassene Tmar

Video Transcoding Product Lead, Intel


Hassene Tmar Received his B.S. degree in Computer Engineering from the University of British Columbia in 2013. He joined eBrisk Video in 2013, where he developed video quality improvement algorithms for eBrisk’s HEVC-based real-time encoders, led in the integration of the FPGA-implemented expanded motion estimation module with the rest of the encoders, and managed the testing and benchmarking experiments for such encoders.

Once at Intel, Hassene led a team to Open Source the Scalable Video Technology codecs (namely, SVT-HEVC, SVT-VP9 and SVT-AV1), optimize such encoders on Intel Xeon CPUs, and provide extensive technical support to Intel’s customers towards the deployment of next-generation video transcoding products based on the Open Source SVT encoders.

Conference Talk

SVT-AV1 Overview, Latest Performance Results and Roadmap

Intel has recently launched the open visual cloud, which will address the growing cloud media markets through the development and optimization of visual components and solutions for visual cloud workloads in five major market segments, namely, media processing and delivery, media analytics, immersive media, cloud graphics and cloud gaming

Such visual cloud workloads can be represented by four major building blocks: Decode, Inference, Render and Encode. Perhaps the most costly block is the Encode block, particularly when implemented in software.

Scalable Video Technology (SVT) is a standard-agnostic architecture that allows software encoders to scale efficiently using a multi-core Xeon™ CPU, the main processing unit used in cloud environments. To address the very-high complexity of the AV1 encoder, Intel and Netflix have recently partnered to develop SVT-AV1, a scalable implementation of the AV1 standard, to the open source community under a permissive BSD+Patent license, and they are currently working on completing the implementation of the features, while targeting both the VOD and Live applications.

SVT-AV1 has evolved rapidly since it was open sourced, and it is now capable of real-time 1080/4K encoding while maintaining high levels of video quality and offering great bandwidth reduction advantages relative to both AVC and HEVC encoders. In this presentation, we will discuss briefly the SVT architecture, review the supported AV1 features through some SVT-implementation examples, and present the latest performance-quality results. We will also present the short-term and long-term roadmaps for SVT-AV1, and discuss some of the outstanding challenges.