Candidate box processing method and device, equipment and medium

Abstract

The embodiment of the invention provides a candidate box processing method and device, equipment and a medium, which are applied to a target detection network; the target detection network comprises acandidate box generation module and a target detection module, and the method comprises the following steps: predicting the preference degree of the target detection module to each candidate box output by the candidate box generation module; according to the current use stage of the target detection network, inputting candidate boxes with corresponding preference degrees in the candidate boxes output by the candidate box generation module into the target detection module, wherein in different use stages, the preference degrees of the candidate boxes input into the target detection module aredifferent.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a candidate frame processing method, device, equipment and medium. Background technique [0002] Object Detection is a core technology of computer vision. Taking the object as an example, object detection aims to locate each object in the image while giving the correct object category. Object detection as a basic vision algorithm has a wide range of applications. [0003] In related technologies, the target detection algorithm mainly adopts a two-stage detection method, which usually needs to generate candidate frames first, and then further classify and reposition the candidate frames. However, the generation of candidate boxes and the classification and relocalization of candidate boxes in the current two-stage detection method are two separate stages from each other. That is, the generation and re-detection of candidate boxes often exist as separate subtasks...

AI Technical Summary

Problems solved by technology

However, the generation of candidate boxes and the classification and relocation of candidate boxes in the current two-stage detection method are two stages separated from each other.

That is, the generation and re-detection of candidate frames often exist as separate sub-tasks, which leads to low efficiency and low accuracy of target detection

Images