SVT-AV1 adds VMAF tuning without runtime VMAF evaluation
The Alliance for Open Media's SVT-AV1 project has merged a new `TUNE-VMAF` encoding mode, which optimizes video quality for VMAF without real-time VMAF model execution. This update introduces preprocessing techniques like luma unsharp masking, adaptive sharpening, noise-aware modulation, and QP-adaptive delta clipping.
Key Takeaways
- `--tune 5` adds TUNE-VMAF to SVT-AV1.
- The new mode uses a per-frame luma unsharp mask built from a cascaded separable box blur.
- Adaptive sharpening strength uses the MIN of per-QP signal and spatial activity.
- Noise-aware modulation is Laplacian-based and reduces sharpening on noisy frames.
- The merge includes AVX2 SIMD optimizations via RTCD for vectorizable steps.
Why It Matters
This gives SVT-AV1 a VMAF-oriented tuning path without the cost of running the VMAF model during encode, which is the central technical change here. The implementation also adds guardrails around the preprocessing stage: noise-aware modulation, QP-adaptive delta clipping, and chroma QP compensation all aim to limit side effects while preserving luma-chroma balance. For streaming workflows, the relevant signal is whether this mode becomes a practical default or remains an opt-in tuning option in later SVT-AV1 releases. Watch for follow-up commits on the M9/M10 optimization work, which the merge request says is still in progress.
Read full article at gitlab.com
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