Title:
NEURAL NETWORK TRAINING METHOD AND APPARATUS FOR IMAGE RETRIEVAL, AND ELECTRONIC DEVICE
Document Type and Number:
WIPO Patent Application WO/2022/032725
Kind Code:
A1
Abstract:
A neural network training method and apparatus for image retrieval, and an electronic device. A neural network comprises a feature extractor and a plurality of learners. The method comprises: regarding each training image group, inputting three images of the training image group into a feature extractor, and determining features of the three images (501); regarding each image in each training image group, respectively multiplying the feature of the image by a random weight corresponding to each learner, so as to obtain a weighted feature corresponding to each learner (502); regarding each image in each training image group, inputting the weighted feature, corresponding to each learner, of the image into the corresponding learner, and determining a plurality of feature vectors of the image (503); and adjusting parameters of a neural network on the basis of the plurality of feature vectors of each image in a plurality of training image groups (504). The influence of information differences between training data on a network is weakened.
Inventors:
REN HAO (CN)
LI SIYANG (CN)
LU HONG (CN)
LI SIYANG (CN)
LU HONG (CN)
Application Number:
PCT/CN2020/111121
Publication Date:
February 17, 2022
Filing Date:
August 25, 2020
Export Citation:
Assignee:
UISEE TECH ZHEJIANG LTD (CN)
International Classes:
G06K9/62
Foreign References:
CN111242162A | 2020-06-05 | |||
CN111125396A | 2020-05-08 | |||
CN107092661A | 2017-08-25 | |||
CN108021947A | 2018-05-11 | |||
CN110688502A | 2020-01-14 | |||
US20190377930A1 | 2019-12-12 |
Attorney, Agent or Firm:
BEIJING MIZAR STAR INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)
Download PDF: