Skip to main content

AVS3

Help Wanted

This section is in need of contributions. If you believe you can help, please see our Contribution Guide to get started as a contributor!

AVS3 is the most recent codec from the AVS family. It aims to bring a significant coding improvement over AVS2 and HEVC. The specification was divided into 2 phases, the first one (main profile) finished in 2019 and the second (high profile) in 2021.

AVS3 has more flexible coding tree and block partitioning shemes compared to previous generation AVS codecs. It also employs new DCT-VII and DST-VII transforms. The work on AVS3 was sponsored by organizations including Peking University, Pengcheng Laboratory and Huawei.

In 2022 it was added to worldwide DVB standard alongside VVC. As of 2024 there several hardware AVS3 decoders including Mediatek and HiSilicon. We should be seeing more AVS3 in hardware since it was added to DVB toolbox.

uavs3 is an open source high performance software encoder and decoder of AVS3-P2. The project was initialized by the Peking University Shenzhen Graduate School and over the time was optimized for x86 and arm processors. FFmpeg since version 6 supports only avs3 decoding via uavs3d.

Intel partnered with Boya and created SVT-AVS3 which to my knowledge is not available to the public. The only open-source encoder available to general public is uavs3e.