Training image acquisition method, model training method and related devices

Abstract

The invention provides a training image acquisition method applied to object counting, a model training method and a related device, and relates to the field of object counting. The training image acquisition method comprises the following steps: acquiring an initial image set of a target object; wherein the initial image set comprises a plurality of initial images and initial annotations corresponding to the initial images; wherein the initial annotation represents feature information of a target object in the corresponding initial image; extracting graphic information of a target object in each initial image according to the initial label; generating a training image set of the target object according to the graphic information and the at least one scene image; wherein the training imageset comprises a plurality of training images, and each training image comprises graphic information and graphic combination information of any scene image. By using the training image acquisition method provided by the invention, the marking of the training image of the target object can be realized by only needing a small amount of data marking, the marking cost of the training image is effectively reduced, and efficient object counting is realized.

Description

technical field [0001] The present application relates to the field of object counting, and in particular, relates to a training image acquisition method, a model training method and related devices. Background technique [0002] Object counting technology based on machine vision is widely used in crowd density estimation and counting, and is the main technical means for crowd monitoring, crowd monitoring, interest area detection, crowd counting and other applications. The difficulties faced by the object counting task are mainly three points: the target object scale changes greatly, the occlusion is large, and the scene is changeable. [0003] At present, few individuals or institutions have extended the counting technology to the counting applications of other objects. The counting models need to be trained through supervised methods. For each counting target, it is necessary to prepare 10^4 and above labeled pictures. Single target The number of objects labeled is on the...

AI Technical Summary

Problems solved by technology

[0003] At present, few individuals or institutions have extended the counting technology to the counting applications of other objects. The counting models need to be trained through supervised methods. For each counting target, it is necessary to prepare 10^4 and above labeled pictures. Single target The number of objects labeled is on the order of 10^6, and the workload of labeling is huge, so there is a lack of relevant labeling data sets

Images