Convolution operator adaptive selection reasoning acceleration method and device based on memory awareness

The invention discloses a convolution operator adaptive selection reasoning acceleration method and device based on memory perception. The convolution operator adaptive selection reasoning acceleration method based on memory perception comprises the following steps: S101, receiving a convolution operation request for an input image or a feature image; s102, obtaining the available memory of the current system, and finding out a convolution operator which can meet the convolution operation request and has the fastest calculation speed from a preset convolution operator set, wherein the available memory of the system can meet the convolution operation request; and S103, performing convolution operation on the input image or the feature image by adopting the found convolution operator. The invention aims to realize adaptive selection of the convolution operator based on memory perception so as to meet the scheduling strategy requirement of the fastest calculation speed. 本发明公开了一种基于内存感知的卷积算子自适应选择推理加速方法及装置，本发明基于内存感