CodePudding user response:
Conversion network: converts network model can be made of SNPE load of DLC files, can choose quantitative DLC files to run on Hexagon DSP and HAT (note that generally USES 8 bit quantification, and produces in the process of quantitative losses),Load network: instantiate SNPE interface and load DLC files; Select needs runtime (such as CPU, GPU, DSP, HAT, etc.)
Data preparation: prepare model input data (for data preprocessing format conversion, quantitative data, etc.),
Perform network: use SNPE runtime loading and execution model,