Title:
DATA PROCESSING METHOD AND APPARATUS
Document Type and Number:
WIPO Patent Application WO/2024/083121
Kind Code:
A1
Abstract:
A data processing method, which is applied to image processing, such as image segmentation or target detection. The method comprises: acquiring a first image feature corresponding to an image and a text feature corresponding to text; obtaining a plurality of second embedded vectors by means of a neural network according to a plurality of preset first embedded vectors and the first image feature, wherein each second embedded vector corresponds to a candidate region of a target object, and each second embedded vector and the first image feature are fused to obtain a corresponding second image feature; and according to the similarity between the text feature and each of the plurality of second embedded vectors, determining weights respectively corresponding to the second embedded vectors, wherein the plurality of weights are used for performing weighting on the plurality of second image features, such that a predicted region corresponding to the target object is determined. The present application changes the granularity of image features from pixels to a target object, and fuses pixels belonging to the same target as a whole with language codes, such that the processing precision of a model can be improved.
Inventors:
ZHU YI (CN)
LIU JIANZHUANG (CN)
LIU JIANZHUANG (CN)
Application Number:
PCT/CN2023/124977
Publication Date:
April 25, 2024
Filing Date:
October 17, 2023
Export Citation:
Assignee:
HUAWEI TECH CO LTD (CN)
International Classes:
G06F16/33; G06F40/126; G06F40/30; G06N3/0464; G06N3/048; G06N3/084; G06V30/148; G06V30/19; G06V30/41
Foreign References:
| CN115757692A | 2023-03-07 | |||
| CN115115913A | 2022-09-27 | |||
| CN113505193A | 2021-10-15 | |||
| CN113065576A | 2021-07-02 | |||
| CN108776787A | 2018-11-09 | |||
| US20220147715A1 | 2022-05-12 |
Attorney, Agent or Firm:
SHENPAT INTELLECTUAL PROPERTY AGENCY (CN)
Download PDF: