Title:
IMAGE REGION POSITIONING METHOD, MODEL TRAINING METHOD, AND RELATED APPARATUS
Document Type and Number:
WIPO Patent Application WO/2020/182112
Kind Code:
A1
Abstract:
Provided in the present disclosure is an image region positioning method, comprising: according to a candidate image region set in an image to be positioned, generating (102) a region semantic information set; obtaining (103) an enhanced semantic information set corresponding to the region semantic information set by means of a graph convolutional network, the graph convolutional network being used to construct an association between various pieces of region semantic information; by means of an image region positioning network model, obtaining (105) the degree of matching between each piece of enhanced semantic information and a text feature set corresponding to text to be positioned; and determining (106) a target candidate image region from among the candidate image region set according to the degree of matching between the text feature set and each piece of enhanced semantic information. Further disclosed in the present disclosure are a model training method and a related apparatus. In the present disclosure, a graph convolutional network is used to enhance the semantic representation between candidate image regions, thereby being beneficial in improving the accuracy of image region positioning, and improving the capability of understanding an image.
Inventors:
MA LIN (CN)
Application Number:
PCT/CN2020/078532
Publication Date:
September 17, 2020
Filing Date:
March 10, 2020
Export Citation:
Assignee:
TENCENT TECH SHENZHEN CO LTD (CN)
International Classes:
G06T7/246; G06V10/25; G06V10/84
Foreign References:
| CN109903314A | 2019-06-18 | |||
| CN108229287A | 2018-06-29 | |||
| CN108229287A | 2018-06-29 | |||
| CN108228757A | 2018-06-29 | |||
| CN109002852A | 2018-12-14 |
Other References:
See also references of EP 3940638A4
Attorney, Agent or Firm:
BEIJING SAN GAO YONG XIN INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)
Download PDF: