MindSpore Lite Kit on HarmonyOS NEXT: Tiny AI, Big Impact

Read the original article：MindSpore Lite Kit on HarmonyOS NEXT: Tiny AI, Big Impact

Introduction

In this article, I will tell you, my dear readers, about the MindSpore lite kit. Today, AI applications aren’t limited to powerful servers. Fast, secure and offline AI experiences are made possible by on-device inference solutions that work in mobile and embedded systems. The MindSpore lite kit is a lightweight AI engine developed for just that need and works integrated with the HarmonyOS NEXT operating system.

Based solely on the official document, we will discuss in detail what the MindSpore lite kit offers, in which scenarios it can be used, its advantages, and the development process.
Let's get started

What is the MindSpore lite kit?

MindSpore is a lightweight AI inference engine embedded in HarmonyOS NEXT. This open-source AI framework provides a friendly development environment for data scientists, algorithm engineers, and developers. Its multi-core architecture supports high-performance inference processes and aims to strengthen the software-hardware ecosystem.

Use Scenarios

The MindSpore lite kit is currently being used successfully in a variety of visual AI tasks:

Image Classification: For example, identifying an image as a cat, dog, plane, or car.
Object Recognition: Recognize, label, and mark objects in camera view with bounding boxes.
Image Segmentation: Positioning a particular object in an image at the pixel level.

Benefits of MindSpore lite

✅ High Performance
It includes core algorithms and optimization techniques that provide high throughput with low latency and low power consumption.

🪶Lightweight Construction
Allows fast operation of small, compressed models. It can work comfortably even on extremely resource-constrained devices.

🌐 Support in All Scenarios
It can work on different operating systems and embedded devices. Offers wide hardware support.

⚙️Efficient Deployment Process
MindSpore, Caffe, and ONNX models are supported. Model conversion, compression, and data manipulation tools are available.

Development Process

There are two basic steps to developing apps with MindSpore lite:

1. Model Conversion

MindSpore lite uses models with a.ms extension for inference. Or convert your existing TensorFlow TFLite, ONNX or Caffe models.

2. Model Distribution

The following operations are performed step by step:
Inference media (context) is created according to

device and hardware characteristics.
The model with a .ms extension is installed.
The input data is defined.
Model inference is made and output results are obtained.

Improvement Methods

MindSpore lite can be integrated on HarmonyOS NEXTin two different ways:

Method 1 (Direct Use with ArkTS) : Model execution can be done quickly by providing direct access to MindSpore Lite APIs from the UI layer.

Method 2 (Using Native API) : Algorithm models can be called over ArkTS with N-API by incorporating dynamic libraries.

Relationship to NNRt

MindSpore is compatible with the Neural Network Runtime (NNRt) on HarmonyOS. The NNRt serves as a bridge between the inference engine and the hardware accelerator (e.g., the NPU). Using NNRt, developers can accelerate the AI model on the device’s AI chip.

Conclusion

The MindSpore lite kit provides an efficient, lightweight and powerful AI infrastructure for smart devices integrated with HarmonyOS NEXT. Enables you to develop applications that work in real time and offline with low resource consumption while providing effective solutions for common scenarios such as image processing and object recognition.

If you want to develop AI solutions on HarmonyOS, starting with the MindSpore lite kit is a great step.