Sample selection method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a sample selection method and device, equipment and a storage medium. The sample selection method comprises the following steps: inputting a plurality of to-be-labeled samples into a pre-trained deep neural network model for identification to obtain a target feature corresponding to each to-be-labeled sample and a category corresponding to the target feature; clustering the target features of the same category in the to-be-labeled sample to obtain a plurality of category clusters corresponding to each category; calculating a distance between a target feature corresponding to the same to-be-labeled sample and each clustering center, and taking the distance as a feature distance corresponding to each target feature; calculating a sample score of the to-be-labeled sample based on the plurality of feature distances corresponding to the same to-be-labeled sample; and according to the sample score of each to-be-labeled sample, determining a target labeled sample. The method can effectively improve the accuracy of annotation sample selection.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a sample selection method, device, computer equipment and storage medium. Background technique [0002] The great success of deep neural networks in the field of computer vision benefits from the complex deep network structure and massive labeled data, but in some complex tasks such as detection, segmentation or medical image fields, data labels are often not easy to obtain, and if the Labeling all the sample data obtained requires a lot of labor costs. [0003] At present, active learning technology can obtain samples that are most helpful to the improvement of the current model from a large amount of unlabeled data for labeling, thereby effectively reducing the cost of labeling. However, most of the existing active learning methods are applied to image classification problems. The predicted posterior probability of the model is used to measure the infor...

AI Technical Summary

Problems solved by technology

[0004] Embodiments of the present invention provide a sample selection method, device, computer equipment, and storage medium to solve the problem that the accuracy of labeling sample selection based on active learning technology is not high in current target detection task scenarios

Images