Data labeling method, equipment and device

Abstract

The invention relates to a data labeling method, equipment and device. The method comprises the steps of obtaining general features and a pre-classification result of to-be-labeled data based on a general depth feature model; extracting special features of the to-be-labeled data according to the pre-classification result and a special depth feature model; performing fusion calculation on the general features and the special features to obtain fusion feature information; obtaining a classification annotation prediction result according to the fusion feature information and the special depth feature model; and labeling the to-be-labeled data according to the classification labeling prediction result. By utilizing the data annotation method disclosed by the invention, the data annotation costis greatly reduced and the data annotation efficiency is greatly improved on the premise of ensuring the data annotation quality.

Description

technical field [0001] The present disclosure relates to the technical field of data intelligent identification, and in particular to a data labeling method, device and device. Background technique [0002] Currently, data labeling is basically carried out manually. For example, intelligent driving data labeling is generally manual labeling, and the data generated by each vehicle in intelligent driving is very large every day, resulting in a very high labor cost for the intelligent driving data labeling. In addition, the error rate of manual labeling is relatively high. At present, some data labeling companies will hire part-time personnel to carry out data labeling, which makes the error rate of data labeling higher. In order to control the error rate within a certain range, additional auditors are required to review. Increased labor costs for data labeling. [0003] Although there are open source image recognition databases, such as ImageNet, which can eliminate the manua...

AI Technical Summary

Problems solved by technology

[0002] Now data labeling is basically done manually. For example, intelligent driving data labeling is generally manual labeling, and the data generated by each vehicle in intelligent driving is very large every day, resulting in very high labor costs for intelligent driving data labeling.

In addition, the error rate of manual labeling is relatively high. At present, some data labeling companies will hire part-time personnel to carry out data labeling, which makes the error rate of data labeling higher. In order to control the error rate within a certain range, additional auditors are required to review. Increased labor costs for data labeling

[0003] Although there are open source image recognition databases, such as ImageNet, which can eliminate the manual labeling process, there are many shortcomings, such as too large labeling frames, few labeling methods, and low accuracy, which are difficult to meet the accuracy of image recognition in the field of intelligent driving. and refined requirements

Images