Target object prediction method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a target object prediction method and device, and the method comprises the steps: inputting an original image into a first detection model, and outputting the first position information and rough class information of a target object; cutting the original image to obtain a plurality of target object area images, inputting the plurality of target object area images into a second detection model, and outputting a plurality of pieces of second position information and a plurality of pieces of fine category information of each target object; if the target category object exists, selecting a target prediction object; and generating a position prediction result according to the first position information and the second position information, and taking the fine category information as a category prediction result. According to the embodiment of the invention, the context information of the target object is added, so that the accuracy of the fine category information output from the second detection model is improved. And the target prediction object is selected for the target category object, so that the interference of a non-target prediction object is avoided, and the accuracy of target object prediction is further improved.

Description

technical field [0001] The present invention relates to the technical field of the Internet, in particular to a target object prediction method, device, electronic equipment and computer-readable storage medium. Background technique [0002] As a basic technique in computer vision, target object prediction is generally divided into two subtasks. The first subtask is to determine the location of the target object by judging the foreground and background, and the second subtask is to classify the target object after the location is determined. On the premise of satisfying these two sub-tasks, the prediction model needs to have a strong enough ability to judge the appearance difference and scale difference of the target object after the position is determined. The quality of a predictive model is usually measured by recall and accuracy. Among them, the recall rate reflects the ratio of the number of target objects of a certain category predicted by the prediction model to the...

AI Technical Summary

Problems solved by technology

These methods have obvious advantages in reducing the false alarm rate and improving the accuracy rate. However, for scenarios where the prediction model needs to ensure high recall, the online difficult sample mining method only retains samples with a higher loss function, completely ignoring the simple ones. Samples, which essentially change the input distribution during training (only contain difficult samples), will cause the prediction model to lose the ability to discriminate easy samples during learning, and cannot fully guarantee the recall of simple samples

For the method designed at the loss function level, the setting of parameters will play a decisive role in the learning of positive and negative samples, which is not conducive to stable result output

For the scheme of collecting negative samples for offline learning of the prediction model, the judgment of the inter-class error of the prediction model will be increased. Although most of the negative samples can be identified, but because the negative samples themselves have a high appearance similarity with the positive samples, In the absence of contextual reference, the induced inter-class error reduces recall

On the other hand, the negative samples mined are limited and can only meet the false detection target of the current training set, and the generalization ability is insufficient.

Images