Performance detection method and device for image recognition model, server and storage medium

Abstract

The invention provides a performance detection method and device for an image recognition model, a server and a storage medium, and belongs to the technical field of the Internet. The method comprisesthe following steps: obtaining a set of test samples, wherein the test sample set comprises a plurality of test samples and a plurality of corresponding real labels; respectively inputting the plurality of test samples into a to-be-detected image recognition model, acquiring a plurality of prediction tags output by the image recognition model, acquiring a plurality of text similarities, using onetext similarity to represent the similarity between the real tag and the prediction tag of one test sample, and determining the recognition accuracy of the image recognition model according to the plurality of text similarities. The recognition accuracy of the model is determined according to the plurality of text similarities, the method can be compatible with label systems of various image recognition models, the performance detection of the model can be realized without re-labeling a sample, the consumption of time and cost is greatly reduced, and the efficiency of the performance detection of the model is improved.

Description

technical field [0001] The present disclosure relates to the technical field of the Internet, and in particular to a performance detection method, device, server and storage medium of an image recognition model. Background technique [0002] With the development of Internet technology, people rely more and more on obtaining various information through the Internet. For example, after taking a picture of an item that one does not know, image recognition is performed through the network to obtain the name of the item. Among them, image recognition can be realized by using various image recognition models, and in order to ensure online experience, image recognition models need to perform performance testing on the model before being applied to the online real environment. [0003] The practice of related technologies is to select multiple pictures from the online real environment as test samples, mark each test sample with a real label according to the label system of the imag...

AI Technical Summary

Problems solved by technology

[0005] Embodiments of the present disclosure provide a performance detection method, device, server, and storage medium of an image recognition model, which can solve the problem of testing the performance of image recognition models with different label systems in the related art. Re-labeling of samples takes a lot of time and cost, and the performance detection efficiency of the model is low

Images