Title:
NEURAL NETWORK TRAINING METHOD AND APPARATUS, CHARACTER RECOGNITION METHOD AND APPARATUS, DEVICE, STORAGE MEDIUM, AND COMPUTER PROGRAM PRODUCT
Document Type and Number:
WIPO Patent Application WO/2023/040626
Kind Code:
A1
Abstract:
A neural network training method and apparatus, a character recognition method and apparatus, a device, a storage medium, and a computer program product. The neural network training method comprises: acquiring training data (S101); performing extraction on the training data on the basis of a first feature extraction layer of a first neural network to obtain a first feature map, and performing extraction on the training data on the basis of a second feature extraction layer of a second neural network to obtain a second feature map, wherein the network precision of the first neural network is higher than the network precision of the second neural network (S102); determining a loss function value of the second neural network on the basis of the first feature map and the second feature map (S103); and adjusting the second neural network on the basis of the determined loss function value to obtain a trained second neural network (S104). The method can improve training precision while ensuring training efficiency.
Inventors:
ZHANG ZHENGFU (CN)
LIANG DING (CN)
WU YICHAO (CN)
LIANG DING (CN)
WU YICHAO (CN)
Application Number:
PCT/CN2022/114914
Publication Date:
March 23, 2023
Filing Date:
August 25, 2022
Export Citation:
Assignee:
SHANGHAI SENSETIME INTELLIGENT TECH CO LTD (CN)
International Classes:
G06N3/08
Foreign References:
| CN113780556A | 2021-12-10 | |||
| CN113378866A | 2021-09-10 | |||
| CN111414987A | 2020-07-14 | |||
| CN111325318A | 2020-06-23 | |||
| CN112288086A | 2021-01-29 | |||
| US20180068218A1 | 2018-03-08 |
Attorney, Agent or Firm:
CHINA PAT INTELLECTUAL PROPERTY OFFICE (CN)
Download PDF: