Image processing method, device and computer-readable storage medium

Abstract

The embodiment of the present invention discloses an image processing method, device, and computer-readable storage medium; in the embodiment of the present invention, after determining the batch data that currently needs to be trained from the training sample set, the batch data includes a plurality of image samples, and the pre-set Let the image processing model perform feature extraction on image samples to obtain a set of feature maps, pruning the feature maps in the feature map set to obtain a set of feature maps after pruning, and perform batch regression on the batch data according to the set of feature maps after pruning Unification processing, to converge the image processing model, return to the step of determining the current batch data to be trained from the training sample set, until the image processing model converges, and obtain the trained image processing model, based on the trained image processing model to process the image The processing is performed to obtain a processing result; this scheme can improve the accuracy of the image processing model for image processing.

Description

technical field [0001] The present invention relates to the field of communication technology, in particular to an image processing method, device and computer-readable storage medium. Background technique [0002] In recent years, with the great popularity of neural network technology in the field of artificial intelligence, the application of neural network to image processing has also made great progress. Especially the application of neural networks to the field of image classification and segmentation. In order to improve the accuracy of the neural network for image processing, it is necessary to improve the network performance during the image processing model training process, and network regularization is an important algorithm to improve network performance. The existing dropout (a network pruning algorithm) is the main The representative network pruning method is widely used in various image processing tasks. [0003] In the process of research and practice of th...

AI Technical Summary

Problems solved by technology

[0003] In the process of research and practice of the prior art, the inventors of the present invention found that the mean value and variance calculated by the batch normalization layer in the image processing model will be inaccurate after the pruning is completed in the existing network pruning method, A variance shift is generated, therefore, it will greatly affect the accuracy of the image processing model for image processing

Images