Image target detection model training method and device based on consistency negative sample

Abstract

The invention relates to the technical field of image processing, in particular to an image target detection model training method and device based on consistency negative samples. In order to solve the problem that in the prior art, a fixed negative sample is adopted, so that a model difficultly learns features with discrimination force, the invention provides an image target detection model training method, which comprises the steps of obtaining an initial image sample set based on the overlapping degree of a real frame in a to-be-identified image and a preset initial anchor frame; accordingto the initial image sample set, through a preset image target detection model, obtaining the prediction anchor frame corresponding to the initial anchor frame, and based on the overlapping degree ofthe real frame and the prediction anchor frame, obtaining an updated image sample set; and training the image target detection model through the updated image sample set. By using the method and thedevice of the invention, the image target detection model can be trained by using more comprehensive information.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method and device for training an image target detection model based on consistent negative samples. Background technique [0002] Target detection is to predict the position of all target objects in the image, mark the position of the target object with a rectangular frame, and predict the category of the object in the rectangular frame. At present, most of the robust and efficient methods are based on convolutional neural networks, and the method based on anchor boxes is often used, that is, the anchor boxes are evenly spread on the image at a certain interval in advance. During the prediction process, the model regresses the prior anchor boxes, so that Their shape and position gradually approach the foreground object, and the model predicts the category of the object in the frame to complete the target detection. [0003] In order to supervise the training of the m...

AI Technical Summary

Problems solved by technology

[0007] In order to solve the above-mentioned problems in the prior art, that is, in order to solve the problem that the model is difficult to learn discriminative features due to the use of fixed negative samples in the prior art, the first aspect of the present invention provides an image based on consistent negative samples A target detection model training method, the method comprising:

Images