US 12,148,215 B2
Systems, methods, and media for main group identification in images via social relation recognition
Kevin James Cannons, Vancouver (CA); Zhan Xu, Richmond (CA); and Minlong Lu, Markham (CA)
Assigned to HUAWEI TECHNOLOGIES CO., LTD., Shenzhen (CN)
Filed by Kevin James Cannons, Vancouver (CA); Zhan Xu, Richmond (CA); and Minlong Lu, Markham (CA)
Filed on Feb. 4, 2022, as Appl. No. 17/592,790.
Prior Publication US 2023/0252787 A1, Aug. 10, 2023
Int. Cl. G06V 20/50 (2022.01); G06F 18/25 (2023.01); G06N 20/00 (2019.01); G06T 11/00 (2006.01); G06V 10/22 (2022.01); G06V 40/10 (2022.01); G06V 10/82 (2022.01)
CPC G06V 20/50 (2022.01) [G06F 18/25 (2023.01); G06N 20/00 (2019.01); G06T 11/00 (2013.01); G06V 10/225 (2022.01); G06V 40/103 (2022.01); G06T 2210/12 (2013.01); G06V 10/82 (2022.01)] 18 Claims
OG exemplary drawing
 
1. A computer-implemented method for identifying a main group of people in an image, comprising:
processing the image to generate people data identifying a plurality of people at least partially visible in the image;
processing the people data to generate main person data identifying a main person of the plurality of people, each other person of the plurality of people being a secondary person;
processing the people data and the main person data to generate, for each secondary person, relation data characterizing a relation between the secondary person and the main person, wherein the relation data is generated by a model trained using machine learning, wherein the model comprises a trained First Glance network trained using a class-balanced loss;
processing the relation data for the secondary people to generate main group data identifying a main group of people comprising the main person and one or more main group members selected from the secondary people.