Conrad Ho
Conrad Ho
Sep 4, 2018 · 1 min read

Depends on the GOP size, there is frame reference dependency issue across shots or even chucks; and the solution is not discussed here. If long-term reference frame, or “pseudo” background frame is considered for reducing bitrate, the situation is more complex.

In the figure 3, average values are for reference: different videos (or even shots) have very different bitrate-VMAF curve. As the target is to find optimal bitrate for pre-determined VMAF value (or vise versa), if we want to reduce the times of trial encoding to find the sweet point, maybe a low-cost pre-analysis of scene (or shot) type is a possible direction to try? Say, scene classification can be used to filter out impossible bitrate range, or even useless coding tools (say, B-frames is not necessary in static scene, AV1 warping may be useless in rigid motion or complex texture content).

    Conrad Ho

    Written by

    Conrad Ho