Netflix, Inc.
Quantifying encoding comparison metric uncertainty via bootstrapping

Abstract:

In various embodiments, an encoding metric comparison application computes a first set of quality scores associated with a test encoding configuration based on a set of bootstrap quality models. Each bootstrap quality model is trained based on a different subset of a training database. The encoding metric comparison application computes a second set of quality scores associated with a reference encoding configuration based on the set of bootstrap quality models. Subsequently, the encoding metric comparison application generates a distribution of bootstrap values for an encoding comparison metric based on the first set of quality scores and the second set of quality scores. The distribution quantifies an accuracy of a baseline value for the encoding comparison metric generated by a baseline quality model.