US 12,075,104 B2
Quantifying perceptual quality model uncertainty via bootstrapping
Christos Bampis, Los Gatos, CA (US); Zhi Li, Mountain View, CA (US); Lavanya Sharan, Menlo Park, CA (US); Julie Novak, San Jose, CA (US); and Martin Tingley, Los Gatos, CA (US)
Assigned to NETFLIX, INC., Los Gatos, CA (US)
Filed by NETFLIX, INC., Los Gatos, CA (US)
Filed on Mar. 13, 2019, as Appl. No. 16/352,755.
Claims priority of provisional application 62/767,454, filed on Nov. 14, 2018.
Claims priority of provisional application 62/645,774, filed on Mar. 20, 2018.
Prior Publication US 2019/0297329 A1, Sep. 26, 2019
Int. Cl. G06T 7/00 (2017.01); G06F 18/214 (2023.01); G06N 20/20 (2019.01); G06V 10/774 (2022.01); H04N 19/154 (2014.01); H04N 21/25 (2011.01); H04N 17/00 (2006.01); H04N 19/147 (2014.01)
CPC H04N 21/252 (2013.01) [G06F 18/214 (2023.01); G06N 20/20 (2019.01); G06T 7/0002 (2013.01); G06V 10/774 (2022.01); H04N 19/154 (2014.11); G06T 2207/10016 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/30168 (2013.01); H04N 17/004 (2013.01); H04N 19/147 (2014.11)] 9 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
performing one or more sampling operations on a training database that includes subjective scores to generate a plurality of resampled datasets;
for each resampled dataset, performing one or more machine learning operations based on the resampled dataset to generate a different bootstrap perceptual quality model;
generating a plurality of perceptual quality scores for a portion of encoded video content using the different bootstrap perceptual quality models; and
performing one or more operations based on the bootstrap perceptual quality models to generate a first confidence interval from a subset of the plurality of perceptual quality scores that meets a first threshold criterion, wherein the first confidence interval quantifies an quantify the accuracy of a perceptual quality score generated by a baseline perceptual quality model for the portion of encoded video content, wherein the baseline perceptual quality model does not comprise one of the bootstrap perceptual quality models, and wherein the portion of encoded video content is not included in the training database.