US 11,810,385 B2
Subject identification based on iterated feature representation
Dongdong Chen, Bellevue, WA (US); Dengpan Fu, Beijing (CN); Jingdong Wang, Beijing (CN); and Jianmin Bao, Beijing (CN)
Assigned to Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed by Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed on Dec. 28, 2020, as Appl. No. 17/135,315.
Prior Publication US 2022/0207264 A1, Jun. 30, 2022
Int. Cl. G06V 40/10 (2022.01); G06V 20/20 (2022.01); G06F 18/22 (2023.01); G06F 18/25 (2023.01)
CPC G06V 40/10 (2022.01) [G06F 18/22 (2023.01); G06F 18/25 (2023.01); G06V 20/20 (2022.01)] 18 Claims
OG exemplary drawing
 
1. A computer-vision method, comprising:
recognizing, at a computing system, a feature representation of a query image depicting an unknown subject to be identified;
computing a similarity score between the feature representation of the query image and a feature representation of each of a plurality of gallery images, the plurality of gallery images collectively depicting two or more different subjects with at least two or more gallery images for each different subject, and each gallery image having a label identifying which of the two or more subjects is depicted by the gallery image;
sequentially iterating one or more updated feature representations of the query image based on one or more of the computed similarity scores;
iterating one or more updated feature representations of one or more gallery images of the plurality of gallery images;
for each of the one or more updated feature representations of the query image, computing an updated similarity score between the updated feature representation of the query image and the updated feature representations of each of the one or more gallery images of the plurality of gallery images to give one or more updated similarity scores; and
computer-identifying the unknown subject as a selected subject of the two or more subjects based on a gallery image labeled as depicting the selected subject having a highest updated similarity score.