Screening method, device, equipment and readable storage medium for picture training set

Abstract

The embodiment of the present invention discloses a method, device, device and readable storage medium for screening a picture training set, and relates to neural network training technology. The method includes: using the training set to train the target detection network and the semantic segmentation network, and the training set includes Multiple labeled pictures; use the trained target detection network and semantic segmentation network to predict multiple unlabeled pictures, obtain target detection results and semantic segmentation results, and determine the position distribution and prediction of each target in each unlabeled picture At least one of accuracy, area distribution, and size features; from the multiple unlabeled pictures, filter the target pictures whose attribute information of the targets meets the requirements of the network training task, and add the target pictures to the the training set. In this embodiment, a training set that meets the requirements of the network training task is automatically screened out from multiple unlabeled pictures without manual screening, and the flexibility is high.

Description

technical field [0001] Embodiments of the present invention relate to neural network training technology, and in particular, to a method, device, device, and readable storage medium for screening a picture training set. Background technique [0002] The training set is the data sample used to fit the network. When predicting the attribute information of people, vehicles, and objects in pictures through neural networks, a large number of pictures of different environments and scenes are required to form a training set to ensure the accuracy of the neural network. [0003] At present, when training the target detection network and semantic segmentation network, it is necessary to manually browse and screen different pictures, which has many obvious disadvantages: 1) high labor cost and low screening efficiency; 2) need to browse all pictures for screening, resulting in The cost of screening time is high; 3) Manual screening is highly subjective and difficult to form a unified...

AI Technical Summary

Problems solved by technology

[0003] At present, when training the target detection network and semantic segmentation network, it is necessary to manually browse and screen different pictures, which has many obvious disadvantages: 1) high labor cost and low screening efficiency; 2) need to browse all pictures for screening, resulting in The cost of screening time is high; 3) Manual screening is highly subjective and difficult to form a unified standard, resulting in poor reliability of the training set; 4) If the screening standard changes, the already screened training set needs to be readjusted one by one, resulting in poor flexibility

Images