Target detection method, electronic equipment, readable medium and program product

By using multimodal prototype classifiers in the object detection method, using language and visual prototypes for feature matching, and combining traditional classifiers, the problem of low detection accuracy of traditional methods in complex open world scenarios is solved, and high-precision multi-category detection is achieved.

CN119649013APending Publication Date: 2025-03-18FUDAN UNIVERSITY
0 Cites 0 Cited by

Patent Information

Application Number
CN202411869000.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-17
Publication Date
2025-03-18

AI Technical Summary

Technical Problem

Traditional object detection methods are difficult to achieve high-precision detection in complex open-world scenarios, especially when the number of categories surges, with poor generalization ability, making it difficult to distinguish the characteristics of different small categories of objects under the same large category, and may lead to incorrect matching.

Method used

The object detection method based on a multimodal prototype classifier is adopted, and language prototypes and visual prototypes are constructed by obtaining category description information and image collections, which are used to extract and match text features and image features respectively. Combined with traditional classifiers, high-precision recognition of objects in the image to be detected is achieved.

Benefits of technology

Through rich prior information of language modality and visual modality, high-precision detection of various objects in complex open world scenarios is achieved, reducing the number of iterations of model training, and saving computing resources and time costs.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a target detection method, electronic equipment, a readable medium and a program product. According to the method, category vocabularies can be collected in advance, corresponding category description extension information is obtained based on the category vocabularies and used for extracting language prototypes including text features corresponding to object categories, and corresponding reference images are obtained based on the category vocabularies and used for extracting visual prototypes including image features corresponding to the object categories; therefore, the features extracted from the to-be-detected image can be matched with the features contained in the language prototype and the visual prototype respectively, and the category of the object in the to-be-detected image is obtained based on the matching result. Thus, the objects in the to-be-detected image can be classified based on rich prior information of the language modality and the visual modality, and high-precision detection of various objects in a complex scene is realized.
Need to check novelty before this filing date? Find Prior Art