Automatic picture labeling method and system

Abstract

The invention discloses an automatic picture labeling method and system, and the method comprises the following steps: obtaining pictures, putting the pictures into a to-be-classified picture pool, and screening and classifying the pictures; introducing a new picture category and a reference picture according to the demand of the business model; calculating feature vectors of the reference picture and the to-be-classified picture; calculating the similarity between the to-be-classified pictures and the reference pictures, and automatically classifying the to-be-classified pictures to the category of the reference picture with the highest similarity; checking automatic classification results of the machine in batches to obtain a confirmed picture data set; and training a new model based on the confirmed picture data set. According to the invention, the picture labeling work can be efficiently and accurately completed, the labor cost is greatly reduced, and the landing of artificial intelligence application is accelerated.

Description

technical field [0001] The invention relates to the field of computer deep learning, in particular to a method and system for automatically labeling pictures. Background technique [0002] At present, there are more and more artificial intelligence applications, but deep learning requires a large amount of high-quality labeled data, which is a very time-consuming task. At present, the most commonly used method is to recruit a large number of labelers or outsource to labeling companies for labeling, which is costly. [0003] Patent CN202010385771 (an image data labeling method), which distributes the labeling, review and management of pictures to labelers, reviewers and administrators respectively, and coordinates the work of all parties through labeling software, but essentially still relies on manpower for labeling , has been unable to meet the growing demand for labeling. [0004] Patent CN201810400584.X (an image annotation system based on crowdsourcing), which crowdsou...

AI Technical Summary

Problems solved by technology

[0002] At present, there are more and more applications of artificial intelligence, but deep learning requires a large amount of high-quality labeled data, which is a very time-consuming task

At present, the most commonly used method is to recruit a large number of labelers or outsource to labeling companies for labeling, which is costly

[0003] Patent CN202010385771 (an image data labeling method), which distributes the labeling, review and management of pictures to labelers, reviewers and administrators respectively, and coordinates the work of all parties through labeling software, but essentially still relies on manpower for labeling , can no longer meet the growing demand for labeling

[0004] Patent CN201810400584.X (an image annotation system based on crowdsourcing), which crowdsources the annotation work to volunteers. Although the annotation cycle can be shortened through multi-person collaboration, it does not reduce the workload of annotation. Time-consuming

[0005] Patent CN202010355551 (zero-shot image recognition method based on attribute feature vector and reversible generative model), which uses a known category of image data sets to train a reversible generative model, uses this model to generate new image data, and trains an SVM classifier accordingly, but As can be seen in step S1-2, the step of manual labeling is still required

Images