Overview

Before You Start

This section aims to provide developers with a quick understanding of the general process of model porting, training, and loading for inference based on CANN through a simple practice.

Quick start operations will be explained using the Atlas 800 training server (model 9010) as an example. If you use the Ascend AI Processor of another model, the supported functions and operations may differ.

Environment Setup

• Install the driver, firmware, and CANN Toolkit by referring to CANN Software Installation Guide.
• Install the PyTorch framework, torch_npu plugin, and PyTorch extension library torchvision by referring to Ascend Extension for PyTorch Configuration and Installation .

Functions

Image classification is the most basic application of computer vision. Given an image, you can determine the class to which the image belongs, for example, cat, dog, airplane, or car.

This practice mainly ports the training script of a ResNet-50 classification network in the PyTorch framework to the Ascend platform for training, saves the trained model and converts it to an offline OM model adapted to the Ascend platform, and finally loads and executes this OM model through the AscendCL application to implement the most basic image classification AI application.