Title:
NETWORK TRAINING METHOD, IMAGE SEGMENTATION METHOD, APPARATUS, DEVICE, MEDIUM, AND PRODUCT
Document Type and Number:
WIPO Patent Application WO/2022/127071
Kind Code:
A1
Abstract:
Provided are a network training method, image segmentation method, apparatus, device, medium, and product. The network training method comprises: according to a first training set, training first and second segmentation networks in an initial state to obtain first and second segmentation networks in a first state, the first training set comprising a labeled first sample image and an unlabeled second sample image; labeling the unlabeled second sample image by means of a second segmentation network in a first state and a trained shape constrained network to determine third and fourth labeled sample images, the labeling accuracy of the third sample image being greater than the labeling accuracy of the fourth sample image; according to the second training set, training the first segmentation network in the first state to obtain a first segmentation network in a target state, the second training set comprising the labeled first, third, and fourth sample images.
Inventors:
WANG WENJI (CN)
XIA QING (CN)
HU ZHIQIANG (CN)
XIA QING (CN)
HU ZHIQIANG (CN)
Application Number:
PCT/CN2021/103070
Publication Date:
June 23, 2022
Filing Date:
June 29, 2021
Export Citation:
Assignee:
SHANGHAI SENSETIME INTELLIGENT TECH CO LTD (CN)
International Classes:
G06N3/08; G06T7/11
Foreign References:
| CN112541928A | 2021-03-23 | |||
| CN109871883A | 2019-06-11 | |||
| CN111598900A | 2020-08-28 | |||
| US10025950B1 | 2018-07-17 |
Attorney, Agent or Firm:
CHINA PAT INTELLECTUAL PROPERTY OFFICE (CN)
Download PDF:
Previous Patent: ENCAPSULATING COMPOSITION, COMPOSITION, ENCAPSULATING GLUE FILM, ELECTRONIC COMPONENT AND SOLAR CELL...
Next Patent: OPTICAL MODULE
Next Patent: OPTICAL MODULE