Title:
TEXT AND VIDEO CROSS-SEARCHING METHOD AND APPARATUS, MODEL TRAINING METHOD AND APPARATUS, DEVICE, AND MEDIUM
Document Type and Number:
WIPO Patent Application WO/2024/098524
Kind Code:
A1
Abstract:
The embodiments of the present invention disclose a model training method and apparatus for cross-searching between video data and text data, a cross-searching method and apparatus between the video data and the text data, a cross-searching device and a nonvolatile readable storage medium, which are applied to information retrieval technologies. The method comprises: for each group of training samples in a training sample set, generating a text graph neural network by means of taking node features corresponding to current sample text data as the node features and taking an inclusion relationship among the node features as a connection relationship; generating a video graph neural network on the basis of taking features of each frame image in an image sequence feature of target sample video data as the node features, as well as on the basis of an edge connection relationship determined by a correlation between the features of each frame image; training a cross-searching model by means of using sample text features which merge third-type text data features and second-type text data features extracted by the text graph neural network, and by using sample video features extracted by the video graph neural network.
More Like This:
Inventors:
LI RENGANG (CN)
WANG LI (CN)
FAN BAOYU (CN)
GUO ZHENHUA (CN)
WANG LI (CN)
FAN BAOYU (CN)
GUO ZHENHUA (CN)
Application Number:
PCT/CN2022/141679
Publication Date:
May 16, 2024
Filing Date:
December 23, 2022
Export Citation:
Assignee:
SUZHOU METABRAIN INTELLIGENT TECH CO LTD (CN)
International Classes:
G06F16/332
Attorney, Agent or Firm:
KANGXIN PARTNERS, P.C. (CN)
Download PDF: