Convolution calculation method, convolution calculation device and terminal equipment

Abstract

The invention is suitable for the technical field of deep learning, and provides a convolution calculation method, a convolution calculation device, terminal equipment and a computer readable storagemedium, and the method comprises the steps: inputting a to-be-processed image into a deep learning model, and obtaining a to-be-blocked convolution group and a blocked target size from all convolutionlayers of the deep learning model; according to the target size, blocking all input channel data of a first to-be-blocked convolution layer in the to-be-blocked convolution group, and the size of each block being the target size; according to all blocks of all input channel data of the first to-be-blocked convolution layer, obtaining an output result of the to-be-blocked convolution group; and inputting an output result of the convolution group to be partitioned into a specified network of the deep learning model. According to the method, the bandwidth consumption can be adjusted by adjustingthe partitioning size of the convolution layer to be partitioned, and frequent updating and upgrading of the deep learning model are self-adapted.

Description

technical field [0001] The present application belongs to the technical field of deep learning, and in particular relates to a convolution calculation method, a convolution calculation device, a terminal device, and a computer-readable storage medium. Background technique [0002] Deep learning is to learn the internal laws and representation levels of sample data. The information obtained during these learning processes is of great help to the interpretation of data such as text, images and sounds. A deep learning model usually includes a convolutional layer. One of the key points of the convolutional computing efficiency of the convolutional layer is how to save data handling and power consumption. If the data multiplexing is not done well, it is easy to form a bandwidth bottleneck. For the existing convolution calculation method, once the design is completed, the data reuse method cannot be adjusted, and the power consumption is also determined accordingly. In this way, d...

AI Technical Summary

Problems solved by technology

Deep learning models usually include a convolutional layer. One of the key points of the convolutional computing efficiency of the convolutional layer is how to save data handling and power consumption. If the data reuse is not done well, it is easy to form a bandwidth bottleneck.

In the existing convolution calculation method, once the design is completed, the data reuse method cannot be adjusted, and the power consumption is also determined accordingly. In this way, different deep learning models will have different efficiency performances, and it is difficult to adapt to the frequent deep learning models. Update status

Images