HARDWARE ENVIRONMENT-BASED DATA QUANTIZATION METHOD AND APPARATUS, AND READABLE STORAGE MEDIUM

摘要

A hardware environment-based data quantization method and apparatus, and a computer readable storage medium. The method comprises: parsing a model file under the current deep learning framework to obtain intermediate calculation graph data and weight data that are irrelevant to a hardware environment, and performing intermediate calculation graph process calculation on image data in an input data set to obtain feature graph data; separately performing uniform quantization on the weight data and each layer of feature graph data according to a preset linear quantization method, and performing calculation to obtain a weight quantization factor and a feature graph quantization factor (S103); combining the weight quantization factor and the feature graph quantization factor to obtain a quantization parameter that makes hardware use shift to replace division; and finally, according to a hardware requirement, writing the quantization parameter and the quantized weight data to a bin file so as to generate quantized file data (S105). Therefore, the present invention solves the problems in the related art of quantization software package redundancy and dependency library conflict caused by supporting a plurality of deep learning frameworks.