Part 11/15:

The goal: maximize compression while preserving perceived quality.

Creating Multiple Encodings for Best Results

To refine compression settings, the speaker recommends generating several encoded variants at different parameter levels—such as different Constant Rate Factors (CRF)—and then evaluating them perceptually. Because no single metric perfectly correlates with human perception, tools like VMAF (Video Multimethod Assessment Fusion) are invaluable.

VMAF, developed by Netflix, combines multiple metrics and trained machine learning models to predict perceived quality. It involves:

• Comparing the original video with various compressed versions.

• Incorporating human observer data.

• Producing a perceptual quality score to inform parameter tuning.