AV2 Video Codec Architecture, presented by Andrey Norkin, Netflix

About this video

• Video Title: AV2 Video Codec Architecture
• Channel: Alliance for Open Media
• Speakers: Andrey Norkin
• Duration: 00:25:54

Overview

This video presents the architecture of the AV2 video codec, detailing its advancements over AV1. Andrey Norkin, a research scientist at Netflix and co-chair of the codec working group at the Alliance for Open Media, explains the various tools and features incorporated into AV2, highlighting improvements in compression efficiency, prediction, transforms, quantization, and loop filters. The presentation also touches upon specific use cases like screen content coding and multi-layer video.

Key takeaways

• AV2 Development Status: The AV2 codec development is nearing completion, with low-level tools finalized and specification writing ongoing. A final release is anticipated by the end of the year.
• Performance Improvements: AV2 demonstrates significant gains in compression efficiency, achieving approximately 28.6% PSNR YUV BD-rate savings and 32.6% VMAF bit-rate savings compared to AV1.
• Core Architecture: AV2 maintains a hybrid block-based model but introduces numerous changes and improvements to tools compared to AV1, including larger block sizes (up to 256x256) and recursive partitioning.
• Advanced Prediction Tools: New and enhanced prediction tools include multi-reference line selection, adaptive intra mode coding, weighted prediction, wedge compound modes, temporal interpolated prediction, optical flow refinement, and warp prediction.
• Transform and Quantization Enhancements: AV2 features include improved transforms with matrix multiplications, learned unitary transforms, and secondary transforms. Quantization has been simplified with an exponential function, and quantization matrices are more flexible and efficient to signal.
• Loop Filters and Restoration: Deblocking has been significantly improved with a generalized formula, and new filters like CCSO, CDEF, and guided detail filter are introduced to enhance picture quality.
• Specific Use Cases: AV2 incorporates features for screen content coding, such as palette mode and improved inter-block copy that allows for in-loop filtering. It also supports multi-layer and multi-stream video, including stereo video.