Parallel convolution method

Abstract

This disclosure proposes a parallel convolution method applicable to a convolution computation circuit. The parallel convolution method includes: dividing an input image into multiple blocks, wherein two adjacent blocks partially overlap with each other; performing convolution operations on the blocks in parallel through multiple cores to obtain multiple sub-feature maps; combining the sub-feature maps to produce a first feature map; inputting the first feature map into a pooling layer to generate a second feature map, wherein the size of the second feature map is smaller than that of the first feature map; and inputting the second feature map into a fully connected layer to obtain at least one output.

Images