HiAI Foundation -- Model Inference

Introduction

This sample code demonstrates how to call HiAI Foundation APIs to perform model inference.

The demo is developed to identify and classify objects in images.

It depends on libhiai_foundation.so, a dynamic library of HiAI Foundation, and libneural_network_core.so, a public dynamic library in the AI domain.

Preview

Home screen	Image classification	Next image

Instructions:

1. Touch "hiaiDemo" on the phone home screen to start the demo, and the model will be automatically loaded.
2. Touch Click for classification, and the demo will perform model inference to classify the current image. After the inference is complete, the three most possible classification results are displayed under the image.
3. Touch Click for next image to display the next image.
4. When you exit the demo, the model is automatically uninstalled.

Project Directory

└── entry/src/main                      // Code area
    ├── cpp
    │    ├── types/libentry
    │    │     └── index.d.ts           // API registration file at the native layer
    │    ├── Classification.cpp         // Functions of APIs at the native layer
    │    ├── CMakeLists.txt             // Compilation configurations at the native layer
    │    ├── HIAIModelManager.cpp       // Implementation of the model management class
    │    ├── HIAIModelManager.h         // Definition of the model management class
    ├── ets
    │    ├── entryability 
    │    │     └── EntryAbility.ets      //  Entry point class
    │    ├── pages 
    │    │     └── Index.ets            // Home screen display class
    └── resources
    │    ├── base/media                 // Image resources
    │    │   ├── cup.jpg
    │    │   └── guitar.jpg
    │    ├── rawfile 
    │    │    ├── hiai.om               // Model file
    │    │    └── labels_caffe.txt      // Model output label file

Implementation Details

The sample code uses the following APIs defined in HiAI Foundation and NNCore:

• OH_NN_ReturnCode HMS_HiAIOptions_SetBandMode(OH_NNCompilation* compilation, HiAI_BandMode bandMode)
• HiAI_BandMode HMS_HiAIOptions_GetBandMode(const OH_NNCompilation* compilation)
• OH_NN_ReturnCode HMS_HiAIOptions_SetModelDeviceOrder(OH_NNCompilation* compilation, HiAI_ExecuteDevice* executeDevices, size_t deviceCount);
• HiAI_Compatibility HMS_HiAICompatibility_CheckFromBuffer(const void* data, size_t size);
• OH_NN_ReturnCode OH_NNDevice_GetAllDevicesID(const size_t **allDevicesID, uint32_t *deviceCount);
• OH_NN_ReturnCode OH_NNDevice_GetName(size_t deviceID, const char **name);
• void *OH_NNTensor_GetDataBuffer(const NN_Tensor *tensor);
• OH_NN_ReturnCode OH_NNTensor_GetSize(const NN_Tensor *tensor, size_t *size);
• OH_NN_ReturnCode OH_NNTensor_Destroy(NN_Tensor **tensor);
• OH_NNCompilation *OH_NNCompilation_ConstructWithOfflineModelBuffer(const void *modelBuffer, size_t modelSize);
• void OH_NNCompilation_Destroy(OH_NNCompilation **compilation);
• OH_NN_ReturnCode OH_NNCompilation_SetDevice(OH_NNCompilation *compilation, size_t deviceID);
• void OH_NNCompilation_Destroy(OH_NNCompilation **compilation);
• OH_NN_ReturnCode OH_NNCompilation_Build(OH_NNCompilation *compilation);
• OH_NNExecutor *OH_NNExecutor_Construct(OH_NNCompilation *compilation);
• OH_NN_ReturnCode OH_NNExecutor_GetInputCount(const OH_NNExecutor *executor, size_t *inputCount);
• NN_TensorDesc *OH_NNExecutor_CreateInputTensorDesc(const OH_NNExecutor *executor, size_t index);
• NN_Tensor *OH_NNTensor_Create(size_t deviceID, NN_TensorDesc *tensorDesc);
• OH_NN_ReturnCode OH_NNTensorDesc_Destroy(NN_TensorDesc **tensorDesc);
• OH_NN_ReturnCode OH_NNExecutor_GetOutputCount(const OH_NNExecutor *executor, size_t *outputCount);
• NN_TensorDesc *OH_NNExecutor_CreateOutputTensorDesc(const OH_NNExecutor *executor, size_t index);
• OH_NN_ReturnCode OH_NNExecutor_RunSync(OH_NNExecutor *executor, NN_Tensor *inputTensor[], size_t inputCount, NN_Tensor *outputTensor[], size_t outputCount);
• void OH_NNExecutor_Destroy(OH_NNExecutor **executor);

Call APIs such as OH_NNExecutor_RunSync, pass the image to be classified, and wait for the classification result to display. Then check the logs. See entry/src/main/ets/pages/Index.ets for reference.

Required Permissions

N/A

Dependency

N/A

Constraints

1. This example can run only on standard systems. Supported devices: Huawei phones, tablets, and 2in1.
2. HarmonyOS: HarmonyOS NEXT Developer Beta1 or later
3. DevEco Studio: DevEco Studio NEXT Developer Beta1 or later
4. HarmonyOS SDK: HarmonyOS NEXT Developer Beta1 SDK or later