SVT-AV1 Encode Times: Impact of CDEF and Loop Restoration

This article analyzes how the Constrained Directional Enhancement Filter (CDEF) and Loop Restoration (LR) directly impact encoding times in the SVT-AV1 (libsvtav1) encoder. It examines the computational overhead of these in-loop filters, how they scale across different encoding presets, and the trade-offs between processing speed and visual quality when enabling or disabling them.

Understanding the In-Loop Filters in SVT-AV1

In the AV1 video coding format, CDEF and Loop Restoration are two critical in-loop filtering stages applied after deblocking. Because they run “in-the-loop,” their outputs are used as references for future frames. This means the encoder must perform these calculations in real-time during the encoding process, which heavily impacts CPU utilization and overall encode times.

Direct Impact on Encode Times

In SVT-AV1, CDEF and Loop Restoration are among the most computationally expensive tools. Disabling or restricting them yields a direct, measurable reduction in encode times, though the exact impact depends heavily on the chosen preset.

1. The Overhead of CDEF

The primary bottleneck in CDEF is the direction search algorithm, which requires the encoder to test multiple directions for each block to find the optimal filtering path.

2. The Overhead of Loop Restoration

Loop Restoration is even more computationally demanding than CDEF. It requires intensive pixel-by-pixel mathematical calculations (solving Wiener filter coefficients or applying self-guided filtering) across large frame areas.

3. Combined Impact of Disabling Both Filters

If both CDEF and Loop Restoration are completely disabled, the encoder bypasses two major mathematical bottlenecks.

SVT-AV1 Preset Scaling and Automation

You do not always need to manually disable these filters, as SVT-AV1 automatically adjusts their complexity based on the preset level (-preset 0 to 13):