US 12,299,864 B1
Localized defect detection
Ali Takbiriborujeni, Lynnwood, WA (US); Mehdi Eftekhari Far, Issaquah, WA (US); and Mahesh Viswanathan, White Plains, NY (US)
Assigned to Amazon Technologies, Inc., Reno, NV (US)
Filed by Amazon Technologies, Inc., Reno, NV (US)
Filed on Dec. 9, 2021, as Appl. No. 17/546,807.
Int. Cl. G06K 9/62 (2022.01); G06N 3/08 (2023.01); G06T 7/00 (2017.01); G06V 10/25 (2022.01); G06V 10/778 (2022.01)
CPC G06T 7/0004 (2013.01) [G06V 10/25 (2022.01); G06V 10/7788 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method, comprising:
determining an instance of content includes at least one defect;
analyzing the instance of content, using a set of explainers, to generate a set of potential defect locations;
determining respective confidence values for the potential defect locations of the set of potential defect locations for each of the explainers of the set of explainers, the respective confidence values indicating a likelihood that the potential defect locations contain the at least one defect;
generating respective combined confidence values for each potential defect location of the set of potential defect locations based, at least in part, on a combination of the respective confidence values for each of the explainers of the set of explainers at each potential defect location;
generating, based at least in part on the respective combined confidence values, a visual output of the instance of content, wherein one or more indicators, corresponding to individual combined confidence values, are overlayed over the instance of content at the corresponding potential defect locations to visually represent aggregate results of the respective combined confidence values associated with the one or more indicators; and
providing the visual output to one or more reviewers.