US 12,450,322 B1
System to determine biometric identification training data
Manoj Aggarwal, Seattle, WA (US); Gerard Guy Medioni, Los Angeles, CA (US); Igor Kviatkovsky, Haifa (IL); and Alon Shoshan, Haifa (IL)
Assigned to AMAZON TECHNOLOGIES, INC., Seattle, WA (US)
Filed by AMAZON TECHNOLOGIES, INC., Seattle, WA (US)
Filed on May 24, 2023, as Appl. No. 18/322,785.
Int. Cl. G06F 21/32 (2013.01); G06V 20/70 (2022.01)
CPC G06F 21/32 (2013.01) [G06V 20/70 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A system comprising:
one or more memories, storing first computer-executable instructions; and
one or more hardware processors to execute the first computer-executable instructions to:
determine first image data acquired by an input device;
determine first embedding data using an embedding model to process the first image data;
determine the first embedding data is greater than a first threshold distance from second embedding data in a first embedding space;
based on the first embedding data being greater than the first threshold distance from the second embedding data in the first embedding space, determine a first set of synthetic embedding data that is associated with the first embedding data by one or more of:
generating the first set of synthetic embedding data within a second threshold distance of the first embedding data in the first embedding space; or
selecting the first set of synthetic embedding data based on the first set of synthetic embedding data being within the second threshold distance of the first embedding data within the first embedding space;
determine a first set of synthetic training image data that is associated with the first set of synthetic embedding data; and
train the embedding model using at least the first set of synthetic training image data.