Alliance Open Media (AOMedia), which oversees the development of video coding formats AV1/AV2, the immersive audio format IAMF, and the image format AVIF, je objavil izdanje projekta AVM (github.com) 1.0.0, which develops the official reference implementation of the encoder and decoder for the video coding format AV2. AVM includes optimizations for processors based on ARM64, x86_64, and MIPS architectures, implemented using extended instruction sets NEON, AVX2, SSE2/3/4, MSA, and DSPr2. The AVM code is written in C and distributed under the BSD license.
Codec AV2 (aomedia.org) does not require license fees and is developed as a successor to the AV1 format. Features of the AV2 codec:
- Optimization for streaming applications;
- Improved prediction of inter-frame changes;
- Significant improvements over AV1 in compression performance;
- Enhanced support for virtual and augmented reality capabilities;
- Support for a wider range of visual quality;
- Ability to simultaneously deliver multiple videos within a single video stream with support for separate display on the screen;
- Additional filters for noise suppression, reduction of compression artifacts, and preservation of detail.
When testing one of the experimental releases of the avm library, using the AV2 codec achieved a bitrate reduction of 32.59% compared to the AV1 codec at the same quality level, using quality assessment metrics VMAF (Video Multi-Method Assessment Fusion) developed by Netflix. When using metrics PSRN-YUV (Peak-Signal-to-Noise Ratio 14:1:1), the bitrate was reduced by 28.63%.
Kot v kodeku AV1, je v AV2 vključena hibridna blokovno usmerjena struktura, vendar v nasprotju z AV1 podpira večje superbloke 256×256, popolno rekurzivno razdelitev (partitioning) in bolj učinkovito ločevanje parametrov svetlosti in barve. V AV2 se uporablja enoten eksponencialni kvantizator, ki pokriva širši razpon svetlosti in zagotavlja večjo natančnost kvantizacije za 8-, 10- in 12-bitne videoposnetke ter bolje upravlja nizke bitne hitrosti. Možnosti napovedovanja medkadrnih sprememb so posodobljene za izboljšanje kakovosti modeliranja sprememb svetlosti in barve, upoštevajo pri gradnji modela do 7 prejšnjih kadrov, podpirajo časovno (temporal) interpolacijo in bolje obravnavajo gibanje v videoposnetkih z visoko ločljivostjo ali hitro spreminjajočo se vsebino.
Izvor: linux.org.ru
