US 12,307,749 B2
Method, system and apparatus for training object recognition model
Xudong Zhao, Jiangsu (CN)
Assigned to INSPUR SUZHOU INTELLIGENT TECHNOLOGY CO., LTD., Jiangsu (CN)
Appl. No. 18/012,936
Filed by INSPUR SUZHOU INTELLIGENT TECHNOLOGY CO., LTD., Jiangsu (CN)
PCT Filed Jul. 29, 2021, PCT No. PCT/CN2021/109199
§ 371(c)(1), (2) Date Dec. 25, 2022,
PCT Pub. No. WO2022/052656, PCT Pub. Date Mar. 17, 2022.
Claims priority of application No. 202010956031.X (CN), filed on Sep. 11, 2020.
Prior Publication US 2023/0267710 A1, Aug. 24, 2023
Int. Cl. G06V 10/774 (2022.01); G06F 17/16 (2006.01); G06N 3/084 (2023.01); G06V 10/48 (2022.01); G06V 10/776 (2022.01); G06V 10/82 (2022.01); G06V 10/94 (2022.01)
CPC G06V 10/774 (2022.01) [G06F 17/16 (2013.01); G06N 3/084 (2013.01); G06V 10/48 (2022.01); G06V 10/776 (2022.01); G06V 10/82 (2022.01); G06V 10/955 (2022.01)] 20 Claims
OG exemplary drawing
 
1. A training method for an object recognition model, comprising:
pre-storing a parameter matrix composed of a plurality of feature vectors for representing object feature information into an internal memory;
inputting sample pictures into a deep learning model for object recognition during model training to obtain sample feature vectors for representing feature information of the sample pictures;
extracting the feature vectors corresponding to the sample pictures from the parameter matrix, randomly extracting a certain number of feature vectors from a remaining parameter matrix, and reconstructing all extracted feature vectors to be a new parameter matrix;
multiplying the sample feature vectors and the new parameter matrix to obtain a similarity between each of the sample feature vectors and each feature vector in the new parameter matrix; and
calculating a loss function according to the similarity, performing back propagation of a gradient on the basis of the loss function, updating parameters of the new parameter matrix and the deep learning model, and updating a total parameter matrix in the internal memory on the basis of the updated new parameter matrix to complete this round of training of the deep learning model.