Image data processing method and device, equipment and medium

Abstract

The invention discloses an image data processing method and device, equipment and a medium. The method comprises steps that a to-be-processed image is received; the to-be-processed image is processedby using the image processing convolutional neural network to obtain a processing result of the to-be-processed image; wherein the image processing convolutional neural network is a network obtained by updating an original convolutional neural network based on each fixed-point tensor in the at least one fixed-point tensor and a scaling vector corresponding to each fixed-point tensor; for each fixed-point tensor, the fixed-point tensor and the scaling vector corresponding to the fixed-point tensor are obtained by quantizing each first sub-tensor in a plurality of first sub-tensors of a weight tensor, and the weight tensor is an obtained tensor of the original convolutional neural network and corresponds to the fixed-point tensor; a plurality of first sub-tensors are obtained by decomposingthe weight tensor along a predetermined dimension. According to the method, image data processing precision can be improved on the basis of hardware platform applicability.

Description

technical field [0001] The present invention relates to the field of computers, in particular to an image data processing method, device, equipment and medium. Background technique [0002] Deep learning has enabled unprecedented performance in many computer vision tasks. Although complex deep learning models have good performance, the high storage space and computing resource consumption make it difficult to effectively apply them on various hardware platforms. [0003] Model compression is a commonly used improvement method in the business deployment service of deep learning models, which can greatly improve the reasoning speed of the model. However, due to the different platforms of different business scenarios, the expression methods of model compression on different hardware platforms are not the same. At present, most of the commonly used model compression expression methods are aimed at a single hardware platform or greatly reduce the accuracy of the model in order ...

Images