A deep separable convolutional neural network acceleration method and accelerator

Abstract

The present invention provides a deep separable convolutional neural network acceleration method, including: performing deep convolution on the input neuron, and when performing the deep convolution calculation, the same M lines of the C input channel are in the three-dimensional processing unit PE array Independent and parallel computing, to get the output neurons of the same N rows of the C channel, N

Description

technical field [0001] The present invention relates to the technical field of depth separable convolutional neural network, in particular, to an acceleration method and accelerator for depth separable convolutional neural network. Background technique [0002] Convolutional Neural Networks (CNNs) have seen great performance in areas of computer vision such as image classification and object recognition. CNNs are widely used in autonomous vehicles, IoT devices, and robot vision due to their high accuracy. These applications usually require CNNs to work in an environment with constrained hardware resources and low power consumption. This poses a huge challenge because CNN models usually require millions of parameters and computations, therefore, it is important to design lightweight neural networks. In recent years, there has also been increasing interest in developing small and compact CNN models, which will further help reduce computational requirements. Recent CNN model...

AI Technical Summary

Problems solved by technology

[0003] The main purpose of the present invention is to provide a depth separable convolutional neural network acceleration method and accelerator, so as to solve the technical problem that the depth separable convolution is not optimized in the prior art

Images