Image processing method and device

Abstract

The invention provides an image processing method. The method comprises the steps of acquiring an input feature map; according to a plurality of stored basic convolution kernels and combination parameters, carrying out convolution on an input feature map to obtain an output feature map, wherein the combination parameters are used for indicating the arrangement sequence of standard convolution kernels formed by combining the basic convolution kernels, and the sizes of the basic convolution kernels are smaller than those of the standard convolution kernels; and performing image processing basedon the output feature map to obtain a processing result. According to the image processing method provided by the embodiment of the invention, the storage space of the convolutional neural network model required for image processing can be reduced.

Description

technical field [0001] The present application relates to artificial intelligence, and more specifically, to an image processing method and device in the field of image processing. Background technique [0002] With the continuous development of image processing technology and the continuous improvement of people's requirements for image display quality, convolutional neural networks (CNN) based on deep learning have developed rapidly in the field of image processing, especially in terminal equipment ( For example, there are more and more applications on mobile phones, cameras, smart homes, and self-driving cars, such as face recognition, object detection, and scene segmentation. [0003] Usually, the high-precision convolutional neural network has a large amount of parameters and calculations. Commonly used convolutional neural network models need to occupy hundreds of megabytes of storage space and billions of calculations, while the memory and computing power of terminal ...

AI Technical Summary

Problems solved by technology

[0003] Usually, the high-precision convolutional neural network has a large amount of parameters and calculations. Commonly used convolutional neural network models need to occupy hundreds of megabytes of storage space and billions of calculations, while the memory and computing power of terminal devices Resources are very limited, do not have strong computing power and large cache, making it difficult to deploy high-precision convolutional neural networks on terminal devices

Images