Image processing method, apparatus, storage medium and device

Abstract

The present invention discloses an image processing method, apparatus, storage medium and device, and belongs to the field of deep learning. The method comprises: for each convolution layer of a preset convolutional neural network, acquiring a feature map input in the convolution layer; performing first pre-processing on the feature map, and generating a first matrix according to the pre-processedfeature map, wherein feature data continuously used in the pre-processed feature map is arranged adjacent to the first matrix; performing second pre-processing on the weight of at least one convolution kernel of the convolution layer, and generating a second matrix according to the pre-processed weight, wherein the weights continuously used in the pre-processed weight are arranged adjacent to thesecond matrix; and performing an outer product operation on each row element in the first matrix and the second matrix, and after performing third pre-processing on the obtained outer product operation result, obtaining a convolution operation result output in the convolutional layer. According to the technical scheme of the present invention, in the implementation of the winograd convolution acceleration operation, the computational density and memory access efficiency are effectively improved, and the complexity of hardware implementation is reduced.

Description

technical field [0001] The present invention relates to the field of deep learning technology, in particular to an image processing method, device, storage medium and equipment. Background technique [0002] As a deep learning technology, convolutional neural network has been widely used in the field of image processing. In image processing, after the image to be processed is input into the trained convolutional neural network, the convolutional neural network performs feature extraction on the image to be processed through the convolution operation of multiple convolutional layers, and based on the extracted features, the image to be processed Perform processing such as classification or identification. [0003] As we all know, the complexity of convolution operation is high, so related technologies usually adopt acceleration schemes to complete convolution operations. For example, in view of the winograd convolution method can greatly reduce the complexity of convolution ...

AI Technical Summary

Problems solved by technology

[0005] The embodiment of the present invention provides an image processing method, device, storage medium and equipment, which solves the problem of high requirements on hardware equipment existing in related technologies

Images