The SVT-AV1 4.0.0 (Scalable Video Technology AV1) library has been released, featuring implementations of the AV1 video encoding format encoder and decoder, accelerated by hardware parallel computing extensions present in x86_64 and ARM processors. The project was created by Intel in partnership with Netflix to achieve a performance level suitable for on-the-fly video transcoding and use in video-on-demand (VOD) services. The development is currently underway under the auspices of the Open Media Alliance (AOMedia), which oversees the development of the AV1 video encoding format. The project was previously developed within the OpenVisualCloud project, which also develops the SVT-HEVC and SVT-VP9 encoders. The code is distributed under a BSD license.
SVT-AV1 can be compiled for systems based on any architecture for which there is a compiler supporting the C99 standard, but the best performance is achieved on x86_64 systems, for which assembler optimizations based on SIMD instructions are used (AVX2 support in the CPU is desirable, but SSE2 is sufficient as a minimum). Memory consumption depends on the number of processor cores involved in encoding, controlled by the "--lp" option. Due to the complexity of the algorithms used in AV1, encoding this format requires significantly more resources than other formats, for example, the standard encoder from the AV1 project requires 5721, 5869 and 658 times more computations compared to the x264 (profile "main"), x264 (profile "high") and libvpx-vp9 encoders.
Among the changes in the new SVT-AV1 release:
- Optimization of single-frame and AVIF image encoding modes has been performed. For presets M0-M11, enabling the MS-SSIM parameter (—tune 4) resulted in a 5-8x increase in encoding speed while maintaining quality. When evaluated using the BD-Rate (Bjontegaard-Delta) method, compression efficiency improved by 5-8% at the same complexity level.
- The quality/speed tradeoffs for encoding in "-rtc" (Real-Time Communications) mode have been optimized. For presets M7-M11, a 5-15% speed increase is observed while maintaining quality levels.
- The quality/speed tradeoffs for encoding in Random Access mode have been optimized. For presets M0-M7, a 10-25% speedup is observed without quality loss when using the "—fast-decode=1|2" option.
- On ARM systems, additional optimizations based on the Neon and SVE2 extensions are enabled, resulting in a 5% performance improvement for encoding low-resolution, high-color-depth streams.
- The porting of extended capabilities for the high visual quality mode, enabled by using the "—tune 0" (VQ, Video Quality) settings for video and "—tune 3" (IQ, Image Quality) for Avif images, has been completed from the discontinued SVT-AV1-PSY fork.
- The AC Bias psychovisual modeling method has been implemented, improving the preservation of detail and noise from the camera sensor.
- Improved support for S-frames (Switch Frames), which are used when switching between streams of different quality. S-frame support has been added to the quantization (QP) and decoding order settings.
- Added the ability to use IQ (Image Quality) and MS-SSIM (Multi-Scale Structural Similarity Index) modes when encoding images and individual frames.
- Changes have been made to the API, including those that break backward compatibility.
Source: opennet.ru
