VTM-8.0 adaptive loop filter optimization cuts encoding time by 25%
Researchers from Peking University and the Institute of Computing Technology proposed an optimized Versatile Video Coding (VVC) framework that facilitates parallel execution of GALF and CCALF while minimizing external memory access. The new method achieved approximately 25% computational time-savings for the Adaptive Loop Filter module with negligible impact on coding performance and has been integrated into the VTM-8.0 reference software.
Key Takeaways
- Reduces picture buffer access from 152 passes to a single pass using a new distortion estimation method.
- Achieves approximately 25% computational time-savings for the Adaptive Loop Filter (ALF) module under Random Access configurations.
- Integrates a resolution-dependent linear model to adaptively predict luma filter parameters, reducing redundant RDO calculations.
- Decouples GALF and CCALF dependencies in the encoder, allowing parallel processing previously only possible in decoders.
Why It Matters
The high computational overhead of VVC has historically restricted its use to offline processing, as the ALF module alone requires extensive off-chip memory bandwidth. By slashsing picture buffer access and enabling parallel execution, this optimization removes a significant hurdle for real-time encoder hardware and mobile SoC designs. These efficiency gains maintain original coding performance with negligible BD-rate loss, making high-fidelity 4K and 8K streaming more feasible for low-latency live applications. Watch for further adoption of these techniques into the upcoming open-source VVC reference software (VTM) updates to signal a shift toward broader commercial deployment.
Additional Context
The push for encoder complexity reduction is central to the industry's adoption of the H.266/Versatile Video Coding (VVC) standard. While VVC offers 30% to 50% better compression efficiency than its predecessor, HEVC, its encoding process is up to 31 times more complex, per IEEE reporting from January 2025. This complexity, largely driven by advanced block partitioning and in-loop filtering, has necessitated a surge in research focused on machine learning-based heuristics. For instance, recent surveys in May 2026 indicate that deep learning methods can reduce encoding time by 10% to 55%, though often at the cost of a slight increase in BD-rate. Simultaneously, the Joint Video Experts Team (JVET) has undergone structural changes to facilitate these innovations. Following the 2024 World Telecommunication Standardization Assembly (WTSA), ITU-T consolidated several study groups into the new Study Group 21 (SG21) for the 2025-2028 period. This group continues to oversee the evolution of VVC while exploring generative face video coding and other AI-driven compression techniques through designated Ad hoc Groups. Industry groups like the Media Coding Industry Forum (MC-IF) predicted 2024 to be a breakout year for VVC as it gained approvals from standard bodies like the DVB and ATSC, setting the stage for 2025 and 2026 to focus on practical hardware integration and real-time performance. Energy efficiency and memory bandwidth remain critical roadblocks for VVC on consumer devices. Per research from SigPort, VVC's memory bandwidth requirements are approximately 30 times higher than HEVC for encoders. Innovations like the one-pass CCALF scheme address this by minimizing off-chip data access, which traditionally consumes 10 times more power than on-chip operations. As 8K and immersive media applications scale, the focus within JVET and commercial labs is shifting from pure bitrate reduction to unified rate-distortion-energy optimization to ensure compatibility with mobile power constraints.
Read full article at arxiv.org
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source