Accelerated calculation method of convolutional neural network, storage medium and computer equipment

Abstract

The invention discloses an acceleration calculation method for a convolutional neural network. The method comprises the steps: obtaining a plurality of frames of input data to be processed, at least one group of adjacent frame data exists in the plurality of frames of input data, and the adjacent frame data comprises the previous frame data and the next frame data; performing convolution calculation on the plurality of frames of input data in sequence, when convolution calculation is performed on the next frame of data in the adjacent frame of data, judging whether a time reuse part exists in a convolution window at the same position of the next frame of data and the previous frame of data in the adjacent frame of data, and if yes, performing convolution calculation on the time reuse part; the time reuse parts are same parts at the same position in the convolution window; and if the time reuse part exists, taking the convolution value of the same part at the same position as the time reuse part in the convolution window of the previous frame data as the convolution value of the time reuse part. The time reuse parts and the space reuse parts are recognized in advance, so that repeated calculation is avoided, and the operand is saved.

Description

technical field [0001] The invention belongs to the technical field of data processing, and in particular, relates to an accelerated calculation method of a convolutional neural network, a computer-readable storage medium, and a computer device. Background technique [0002] With the advent of the big data era, compared with traditional machine learning, deep convolutional neural networks with more hidden layers have more complex network structures and stronger feature learning and feature expression capabilities. Since the introduction of deep convolutional neural networks, it has achieved remarkable results in the fields of computer vision, speech recognition, and natural language processing. In order to enhance the accuracy of the neural network, deeper and deeper network structures are designed, but the amount of parameters and calculations increase sharply, which has higher requirements on the data bandwidth and computing power of the hardware platform. [0003] Modern...

AI Technical Summary

Problems solved by technology

[0003] Modern neural network accelerators mainly improve computing power by increasing the operating frequency and increasing the number of computing units. They have already faced problems such as low utilization of computing units and poor scalability. Increasing the operating frequency and increasing the number of computing units will inevitably lead to system performance. increase in consumption

Images