US 11,941,084 B2
Self-supervised learning for anomaly detection and localization
Kihyuk Sohn, Mountain View, CA (US); Chun-Liang Li, Mountain View, CA (US); Jinsung Yoon, San Jose, CA (US); and Tomas Jon Pfister, Foster City, CA (US)
Assigned to Google LLC, Mountain View, CA (US)
Filed by Google LLC, Mountain View, CA (US)
Filed on Nov. 11, 2021, as Appl. No. 17/454,605.
Claims priority of provisional application 63/113,780, filed on Nov. 13, 2020.
Prior Publication US 2022/0156521 A1, May 19, 2022
Int. Cl. G06F 18/214 (2023.01); G06N 3/08 (2023.01); G06V 10/22 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01)
CPC G06F 18/2155 (2023.01) [G06N 3/08 (2013.01); G06V 10/22 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A method for training a machine learning model, the method comprising:
obtaining, at data processing hardware, a set of training samples; and
during each of one or more training iterations, for each training sample in the set of training samples:
cropping, by the data processing hardware, the training sample to generate a first cropped image;
cropping, by the data processing hardware, the training sample to generate a second cropped image that is different than the first cropped image;
duplicating, by the data processing hardware, a first portion of the second cropped image;
overlaying, by the data processing hardware, the duplicated first portion of the second cropped image on a second portion of the second cropped image to form an augmented second cropped image, the first portion different than the second portion; and
training, by the data processing hardware, the machine learning model with the first cropped image and the augmented second cropped image.