Dataset construction method and device, mobile terminal and readable storage medium

Abstract

The invention relates to a dataset construction method and device, a mobile terminal and a computer readable storage medium. The method comprises the steps that a first dataset which contains a firstpreset amount of data and carries annotation information is acquired according to a learning task; a classification model is trained on the first dataset, and precision information of the classification model is assessed; when the precision information reaches a preset value, non-annotated data is screened based on the trained classification model, and the screened data is merged into the first dataset to form a second dataset; and the data in the second dataset is classified and cleaned based on the trained classification model to form a target dataset with a target amount of data. In this way, semi-automatic data collection, screening and annotation can be realized, a large amount of high-quantity data for classification model training can be acquired on the basis of consuming little manpower, manpower cost is greatly saved, and meanwhile dataset construction efficiency is improved.

Description

technical field [0001] The present application relates to the field of computer applications, in particular to a method and device for constructing a data set, a mobile terminal, and a computer-readable storage medium. Background technique [0002] The field of artificial intelligence (AI) is developing rapidly, especially with the wide application of deep learning technology, which has made breakthrough progress in the fields of object detection and recognition. Generally, artificial intelligence AI algorithms are mainly based on supervised learning deep learning technology, and training data is the driving force of artificial intelligence models. [0003] The current training data acquisition methods mainly include open source datasets, web crawling, and offline collection. However, in order to obtain a large amount of data related to learning tasks, it is generally necessary to manually screen, classify and label information on open source datasets and web crawled data. ...

AI Technical Summary

Problems solved by technology

However, in order to obtain a large amount of data related to learning tasks, it is generally necessary to manually screen, classify and label information on open source datasets and web crawled data. It consumes a lot of manpower and material resources, and the cost is very high

Images