US 12,340,566 B2
Multi-view fine-grained identification method, apparatus, electronic device and medium
Zhanyu Ma, Beijing (CN); Kongming Liang, Beijing (CN); Ruoyi Du, Beijing (CN); and Wenqing Yu, Beijing (CN)
Assigned to BEIJING UNIVERSITY OF POSTS AND TELECOMMUNICATIONS, Beijing (CN)
Filed by BEIJING UNIVERSITY OF POSTS AND TELECOMMUNICATIONS, Beijing (CN)
Filed on Apr. 17, 2023, as Appl. No. 18/135,525.
Claims priority of application No. 202210887082.0 (CN), filed on Jul. 26, 2022.
Prior Publication US 2024/0037918 A1, Feb. 1, 2024
Int. Cl. G06V 10/774 (2022.01); G06V 10/764 (2022.01); G06V 10/776 (2022.01)
CPC G06V 10/774 (2022.01) [G06V 10/764 (2022.01); G06V 10/776 (2022.01)] 18 Claims
OG exemplary drawing
 
1. A multi-view fine-grained identification method, comprising;
acquiring a sample data set containing a plurality of multi-view samples, wherein each sample contains a plurality of sub-view images shot from different perspectives, and each sample is marked with a corresponding category;
training an initial fine-grained identification model by using the sub-view images corresponding to each sample respectively to acquire a to-be-optimized fine-grained identification model, wherein the to-be-optimized fine-grained identification model is deployed with an aggregator for aggregating output features and a selector for selecting a next sub-view image as a training object;
optimizing and adjusting the to-be-optimized fine-grained identification model based on classification results of the aggregator and the selector until it is determined that the optimization is completed to acquire a trained target fine-grained identification model; and
performing image identification on the to-be-classified image by using the target fine-grained identification model.