1# 使用MindSpore Lite实现图像分类（C/C++） 2 3<!--Kit: MindSpore Lite Kit--> 4<!--Subsystem: AI--> 5<!--Owner: @zhuguodong8--> 6<!--Designer: @zhuguodong8; @jjfeing--> 7<!--Tester: @principal87--> 8<!--Adviser: @ge-yafang--> 9 10## 场景说明 11 12开发者可以使用[MindSpore](../../reference/apis-mindspore-lite-kit/capi-mindspore.md)，在UI代码中直接集成MindSpore Lite能力，快速部署AI算法，进行AI模型推理，实现图像分类的应用。 13 14图像分类可实现对图像中物体的识别，在医学影像分析、自动驾驶、电子商务、人脸识别等有广泛的应用。 15 16## 基本概念 17 18- N-API：用于构建ArkTS本地化组件的一套接口。可利用N-API，将C/C++开发的库封装成ArkTS模块。 19 20## 开发流程 21 221. 选择图像分类模型。 232. 在端侧使用MindSpore Lite推理模型，实现对选择的图片进行分类。 24 25## 环境准备 26 27安装DevEco Studio，要求版本 >= 4.1，并更新SDK到API 11或以上。 28 29## 开发步骤 30 31本文以对相册的一张图片进行推理为例，提供使用MindSpore Lite实现图像分类的开发指导。 32 33### 选择模型 34 35本示例程序中使用的图像分类模型文件为[mobilenetv2.ms](https://download.mindspore.cn/model_zoo/official/lite/mobilenetv2_openimage_lite/1.5/mobilenetv2.ms)，放置在entry/src/main/resources/rawfile工程目录下。 36 37如果开发者有其他图像分类的预训练模型，请参考[MindSpore Lite 模型转换](mindspore-lite-converter-guidelines.md)介绍，将原始模型转换成.ms格式。 38 39### 编写图像输入和预处理代码 40 411. 此处以获取相册图片为例，调用[@ohos.file.picker](../../reference/apis-core-file-kit/js-apis-file-picker.md) 实现相册图片文件的选择。 42 432. 根据模型的输入尺寸，调用[@ohos.multimedia.image](../../reference/apis-image-kit/arkts-apis-image.md) （实现图片处理）、[@ohos.file.fs](../../reference/apis-core-file-kit/js-apis-file-fs.md) （实现基础文件操作） API对选择图片进行裁剪、获取图片buffer数据，并进行标准化处理。 44 45 ```ts 46 // Index.ets 47 import { fileIo } from '@kit.CoreFileKit'; 48 import { photoAccessHelper } from '@kit.MediaLibraryKit'; 49 import { BusinessError } from '@kit.BasicServicesKit'; 50 import { image } from '@kit.ImageKit'; 51 52 @Entry 53 @Component 54 struct Index { 55 @State modelName: string = 'mobilenetv2.ms'; 56 @State modelInputHeight: number = 224; 57 @State modelInputWidth: number = 224; 58 @State uris: Array<string> = []; 59 60 build() { 61 Row() { 62 Column() { 63 Button() { 64 Text('photo') 65 .fontSize(30) 66 .fontWeight(FontWeight.Bold) 67 } 68 .type(ButtonType.Capsule) 69 .margin({ 70 top: 20 71 }) 72 .backgroundColor('#0D9FFB') 73 .width('40%') 74 .height('5%') 75 .onClick(() => { 76 // 获取相册图片 77 // 1.创建图片文件选择实例 78 let photoSelectOptions = new photoAccessHelper.PhotoSelectOptions(); 79 80 // 2.设置选择媒体文件类型为IMAGE，设置选择媒体文件的最大数目 81 photoSelectOptions.MIMEType = photoAccessHelper.PhotoViewMIMETypes.IMAGE_TYPE; 82 photoSelectOptions.maxSelectNumber = 1; 83 84 // 3.创建图库选择器实例，调用select()接口拉起图库界面进行文件选择。文件选择成功后，返回photoSelectResult结果集。 85 let photoPicker = new photoAccessHelper.PhotoViewPicker(); 86 photoPicker.select(photoSelectOptions, 87 async (err: BusinessError, photoSelectResult: photoAccessHelper.PhotoSelectResult) => { 88 if (err) { 89 console.error('MS_LITE_ERR: PhotoViewPicker.select failed with err: ' + JSON.stringify(err)); 90 return; 91 } 92 console.info('MS_LITE_LOG: PhotoViewPicker.select successfully, photoSelectResult uri: ' + 93 JSON.stringify(photoSelectResult)); 94 this.uris = photoSelectResult.photoUris; 95 console.info('MS_LITE_LOG: uri: ' + this.uris); 96 // 预处理图片数据 97 try { 98 // 1.使用fileIo.openSync接口，通过uri打开这个文件得到fd 99 let file = fileIo.openSync(this.uris[0], fileIo.OpenMode.READ_ONLY); 100 console.info('MS_LITE_LOG: file fd: ' + file.fd); 101 102 // 2.通过fd使用fileIo.readSync接口读取这个文件内的数据 103 let inputBuffer = new ArrayBuffer(4096000); 104 let readLen = fileIo.readSync(file.fd, inputBuffer); 105 console.info('MS_LITE_LOG: readSync data to file succeed and inputBuffer size is:' + readLen); 106 107 // 3.通过PixelMap预处理 108 let imageSource = image.createImageSource(file.fd); 109 if (imageSource == undefined) { 110 console.error('MS_LITE_ERR: createImageSource failed.') 111 return 112 } 113 imageSource.createPixelMap().then((pixelMap) => { 114 pixelMap.getImageInfo().then((info) => { 115 console.info('MS_LITE_LOG: info.width = ' + info.size.width); 116 console.info('MS_LITE_LOG: info.height = ' + info.size.height); 117 // 4.根据模型输入的尺寸，将图片裁剪为对应的size，获取图片buffer数据readBuffer 118 pixelMap.scale(256.0 / info.size.width, 256.0 / info.size.height).then(() => { 119 pixelMap.crop({ 120 x: 16, 121 y: 16, 122 size: { height: this.modelInputHeight, width: this.modelInputWidth } 123 }) 124 .then(async () => { 125 let info = await pixelMap.getImageInfo(); 126 console.info('MS_LITE_LOG: crop info.width = ' + info.size.width); 127 console.info('MS_LITE_LOG: crop info.height = ' + info.size.height); 128 // 需要创建的像素buffer大小 129 let readBuffer = new ArrayBuffer(this.modelInputHeight * this.modelInputWidth * 4); 130 await pixelMap.readPixelsToBuffer(readBuffer); 131 console.info('MS_LITE_LOG: Succeeded in reading image pixel data, buffer: ' + 132 readBuffer.byteLength); 133 // 处理readBuffer，转换成float32格式，并进行标准化处理 134 const imageArr = 135 new Uint8Array(readBuffer.slice(0, this.modelInputHeight * this.modelInputWidth * 4)); 136 console.info('MS_LITE_LOG: imageArr length: ' + imageArr.length); 137 let means = [0.485, 0.456, 0.406]; 138 let stds = [0.229, 0.224, 0.225]; 139 let float32View = new Float32Array(this.modelInputHeight * this.modelInputWidth * 3); 140 let index = 0; 141 for (let i = 0; i < imageArr.length; i++) { 142 if ((i + 1) % 4 == 0) { 143 float32View[index] = (imageArr[i - 3] / 255.0 - means[0]) / stds[0]; // B 144 float32View[index+1] = (imageArr[i - 2] / 255.0 - means[1]) / stds[1]; // G 145 float32View[index+2] = (imageArr[i - 1] / 255.0 - means[2]) / stds[2]; // R 146 index += 3; 147 } 148 } 149 console.info('MS_LITE_LOG: float32View length: ' + float32View.length); 150 let printStr = 'float32View data:'; 151 for (let i = 0; i < 20; i++) { 152 printStr += ' ' + float32View[i]; 153 } 154 console.info('MS_LITE_LOG: float32View data: ' + printStr); 155 }) 156 }) 157 }) 158 }) 159 } catch (err) { 160 console.error('MS_LITE_LOG: uri: open file fd failed.' + err); 161 } 162 }) 163 }) 164 }.width('100%') 165 } 166 .height('100%') 167 } 168 } 169 ``` 170 171### 编写推理代码 172 173调用[MindSpore](../../reference/apis-mindspore-lite-kit/capi-mindspore.md)实现端侧推理，推理代码流程如下。 174 1751. 引用对应的头文件 176 177 ```c++ 178 #include <iostream> 179 #include <sstream> 180 #include <stdlib.h> 181 #include <hilog/log.h> 182 #include <rawfile/raw_file_manager.h> 183 #include <mindspore/types.h> 184 #include <mindspore/model.h> 185 #include <mindspore/context.h> 186 #include <mindspore/status.h> 187 #include <mindspore/tensor.h> 188 #include "napi/native_api.h" 189 ``` 190 1912. 读取模型文件 192 193 ```c++ 194 #define LOGI(...) ((void)OH_LOG_Print(LOG_APP, LOG_INFO, LOG_DOMAIN, "[MSLiteNapi]", __VA_ARGS__)) 195 #define LOGD(...) ((void)OH_LOG_Print(LOG_APP, LOG_DEBUG, LOG_DOMAIN, "[MSLiteNapi]", __VA_ARGS__)) 196 #define LOGW(...) ((void)OH_LOG_Print(LOG_APP, LOG_WARN, LOG_DOMAIN, "[MSLiteNapi]", __VA_ARGS__)) 197 #define LOGE(...) ((void)OH_LOG_Print(LOG_APP, LOG_ERROR, LOG_DOMAIN, "[MSLiteNapi]", __VA_ARGS__)) 198 199 void *ReadModelFile(NativeResourceManager *nativeResourceManager, const std::string &modelName, size_t *modelSize) { 200 auto rawFile = OH_ResourceManager_OpenRawFile(nativeResourceManager, modelName.c_str()); 201 if (rawFile == nullptr) { 202 LOGE("MS_LITE_ERR: Open model file failed"); 203 OH_ResourceManager_CloseRawFile(rawFile); 204 return nullptr; 205 } 206 long fileSize = OH_ResourceManager_GetRawFileSize(rawFile); 207 if (fileSize <= 0) { 208 LOGE("MS_LITE_ERR: FileSize not correct"); 209 } 210 void *modelBuffer = malloc(fileSize); 211 if (modelBuffer == nullptr) { 212 LOGE("MS_LITE_ERR: malloc failed"); 213 } 214 int ret = OH_ResourceManager_ReadRawFile(rawFile, modelBuffer, fileSize); 215 if (ret == 0) { 216 LOGE("MS_LITE_ERR: OH_ResourceManager_ReadRawFile failed"); 217 OH_ResourceManager_CloseRawFile(rawFile); 218 return nullptr; 219 } 220 OH_ResourceManager_CloseRawFile(rawFile); 221 *modelSize = fileSize; 222 return modelBuffer; 223 } 224 ``` 225 2263. 创建上下文，设置线程数、设备类型等参数，并加载模型。本样例模型，不支持使用NNRt推理。 227 228 ```c++ 229 void DestroyModelBuffer(void **buffer) { 230 if (buffer == nullptr) { 231 return; 232 } 233 free(*buffer); 234 *buffer = nullptr; 235 } 236 237 OH_AI_ContextHandle CreateMSLiteContext(void *modelBuffer) { 238 // Set executing context for model. 239 auto context = OH_AI_ContextCreate(); 240 if (context == nullptr) { 241 DestroyModelBuffer(&modelBuffer); 242 LOGE("MS_LITE_ERR: Create MSLite context failed.\n"); 243 return nullptr; 244 } 245 // 本样例模型，不支持配置OH_AI_DeviceInfoCreate(OH_AI_DEVICETYPE_NNRT) 246 auto cpu_device_info = OH_AI_DeviceInfoCreate(OH_AI_DEVICETYPE_CPU); 247 248 OH_AI_DeviceInfoSetEnableFP16(cpu_device_info, true); 249 OH_AI_ContextAddDeviceInfo(context, cpu_device_info); 250 251 LOGI("MS_LITE_LOG: Build MSLite context success.\n"); 252 return context; 253 } 254 255 OH_AI_ModelHandle CreateMSLiteModel(void *modelBuffer, size_t modelSize, OH_AI_ContextHandle context) { 256 // Create model 257 auto model = OH_AI_ModelCreate(); 258 if (model == nullptr) { 259 DestroyModelBuffer(&modelBuffer); 260 LOGE("MS_LITE_ERR: Allocate MSLite Model failed.\n"); 261 return nullptr; 262 } 263 264 // Build model object 265 auto build_ret = OH_AI_ModelBuild(model, modelBuffer, modelSize, OH_AI_MODELTYPE_MINDIR, context); 266 DestroyModelBuffer(&modelBuffer); 267 if (build_ret != OH_AI_STATUS_SUCCESS) { 268 OH_AI_ModelDestroy(&model); 269 LOGE("MS_LITE_ERR: Build MSLite model failed.\n"); 270 return nullptr; 271 } 272 LOGI("MS_LITE_LOG: Build MSLite model success.\n"); 273 return model; 274 } 275 ``` 276 2774. 设置模型输入数据，执行模型推理。 278 279 ```c++ 280 constexpr int K_NUM_PRINT_OF_OUT_DATA = 20; 281 282 // 设置模型输入数据 283 int FillInputTensor(OH_AI_TensorHandle input, std::vector<float> input_data) { 284 if (OH_AI_TensorGetDataType(input) == OH_AI_DATATYPE_NUMBERTYPE_FLOAT32) { 285 float *data = (float *)OH_AI_TensorGetMutableData(input); 286 for (size_t i = 0; i < OH_AI_TensorGetElementNum(input); i++) { 287 data[i] = input_data[i]; 288 } 289 return OH_AI_STATUS_SUCCESS; 290 } else { 291 return OH_AI_STATUS_LITE_ERROR; 292 } 293 } 294 295 // 执行模型推理 296 int RunMSLiteModel(OH_AI_ModelHandle model, std::vector<float> input_data) { 297 // Set input data for model. 298 auto inputs = OH_AI_ModelGetInputs(model); 299 300 auto ret = FillInputTensor(inputs.handle_list[0], input_data); 301 if (ret != OH_AI_STATUS_SUCCESS) { 302 LOGE("MS_LITE_ERR: RunMSLiteModel set input error.\n"); 303 return OH_AI_STATUS_LITE_ERROR; 304 } 305 // Get model output. 306 auto outputs = OH_AI_ModelGetOutputs(model); 307 // Predict model. 308 auto predict_ret = OH_AI_ModelPredict(model, inputs, &outputs, nullptr, nullptr); 309 if (predict_ret != OH_AI_STATUS_SUCCESS) { 310 LOGE("MS_LITE_ERR: MSLite Predict error.\n"); 311 return OH_AI_STATUS_LITE_ERROR; 312 } 313 LOGI("MS_LITE_LOG: Run MSLite model Predict success.\n"); 314 // Print output tensor data. 315 LOGI("MS_LITE_LOG: Get model outputs:\n"); 316 for (size_t i = 0; i < outputs.handle_num; i++) { 317 auto tensor = outputs.handle_list[i]; 318 LOGI("MS_LITE_LOG: - Tensor %{public}d name is: %{public}s.\n", static_cast<int>(i), 319 OH_AI_TensorGetName(tensor)); 320 LOGI("MS_LITE_LOG: - Tensor %{public}d size is: %{public}d.\n", static_cast<int>(i), 321 (int)OH_AI_TensorGetDataSize(tensor)); 322 LOGI("MS_LITE_LOG: - Tensor data is:\n"); 323 auto out_data = reinterpret_cast<const float *>(OH_AI_TensorGetData(tensor)); 324 std::stringstream outStr; 325 for (int i = 0; (i < OH_AI_TensorGetElementNum(tensor)) && (i <= K_NUM_PRINT_OF_OUT_DATA); i++) { 326 outStr << out_data[i] << " "; 327 } 328 LOGI("MS_LITE_LOG: %{public}s", outStr.str().c_str()); 329 } 330 return OH_AI_STATUS_SUCCESS; 331 } 332 ``` 333 3345. 调用以上方法，实现完整的模型推理流程。 335 336 ```c++ 337 static napi_value RunDemo(napi_env env, napi_callback_info info) { 338 LOGI("MS_LITE_LOG: Enter runDemo()"); 339 napi_value error_ret; 340 napi_create_int32(env, -1, &error_ret); 341 // 传入数据处理 342 size_t argc = 2; 343 napi_value argv[2] = {nullptr}; 344 napi_get_cb_info(env, info, &argc, argv, nullptr, nullptr); 345 bool isArray = false; 346 napi_is_array(env, argv[0], &isArray); 347 uint32_t length = 0; 348 // 获取数组的长度 349 napi_get_array_length(env, argv[0], &length); 350 LOGI("MS_LITE_LOG: argv array length = %{public}d", length); 351 std::vector<float> input_data; 352 double param = 0; 353 for (int i = 0; i < length; i++) { 354 napi_value value; 355 napi_get_element(env, argv[0], i, &value); 356 napi_get_value_double(env, value, ¶m); 357 input_data.push_back(static_cast<float>(param)); 358 } 359 std::stringstream outstr; 360 for (int i = 0; i < K_NUM_PRINT_OF_OUT_DATA; i++) { 361 outstr << input_data[i] << " "; 362 } 363 LOGI("MS_LITE_LOG: input_data = %{public}s", outstr.str().c_str()); 364 // Read model file 365 const std::string modelName = "mobilenetv2.ms"; 366 LOGI("MS_LITE_LOG: Run model: %{public}s", modelName.c_str()); 367 size_t modelSize; 368 auto resourcesManager = OH_ResourceManager_InitNativeResourceManager(env, argv[1]); 369 auto modelBuffer = ReadModelFile(resourcesManager, modelName, &modelSize); 370 if (modelBuffer == nullptr) { 371 LOGE("MS_LITE_ERR: Read model failed"); 372 return error_ret; 373 } 374 LOGI("MS_LITE_LOG: Read model file success"); 375 376 auto context = CreateMSLiteContext(modelBuffer); 377 if (context == nullptr) { 378 LOGE("MS_LITE_ERR: MSLiteFwk Build context failed.\n"); 379 return error_ret; 380 } 381 auto model = CreateMSLiteModel(modelBuffer, modelSize, context); 382 if (model == nullptr) { 383 OH_AI_ContextDestroy(&context); 384 LOGE("MS_LITE_ERR: MSLiteFwk Build model failed.\n"); 385 return error_ret; 386 } 387 int ret = RunMSLiteModel(model, input_data); 388 if (ret != OH_AI_STATUS_SUCCESS) { 389 OH_AI_ModelDestroy(&model); 390 OH_AI_ContextDestroy(&context); 391 LOGE("MS_LITE_ERR: RunMSLiteModel failed.\n"); 392 return error_ret; 393 } 394 napi_value out_data; 395 napi_create_array(env, &out_data); 396 auto outputs = OH_AI_ModelGetOutputs(model); 397 OH_AI_TensorHandle output_0 = outputs.handle_list[0]; 398 float *output0Data = reinterpret_cast<float *>(OH_AI_TensorGetMutableData(output_0)); 399 for (size_t i = 0; i < OH_AI_TensorGetElementNum(output_0); i++) { 400 napi_value element; 401 napi_create_double(env, static_cast<double>(output0Data[i]), &element); 402 napi_set_element(env, out_data, i, element); 403 } 404 OH_AI_ModelDestroy(&model); 405 OH_AI_ContextDestroy(&context); 406 LOGI("MS_LITE_LOG: Exit runDemo()"); 407 return out_data; 408 } 409 ``` 410 4116. 编写CMake脚本，链接MindSpore Lite动态库。 412 413 ```c++ 414 # the minimum version of CMake. 415 cmake_minimum_required(VERSION 3.4.1) 416 project(MindSporeLiteCDemo) 417 418 set(NATIVERENDER_PATH ${CMAKE_CURRENT_SOURCE_DIR}) 419 420 if(DEFINED PACKAGE_FIND_FILE) 421 include(${PACKAGE_FIND_FILE}) 422 endif() 423 424 include_directories(${NATIVERENDER_PATH} 425 ${NATIVERENDER_PATH}/include) 426 427 add_library(entry SHARED mslite_napi.cpp) 428 target_link_libraries(entry PUBLIC mindspore_lite_ndk) 429 target_link_libraries(entry PUBLIC hilog_ndk.z) 430 target_link_libraries(entry PUBLIC rawfile.z) 431 target_link_libraries(entry PUBLIC ace_napi.z) 432 ``` 433 434### 使用N-API将C++动态库封装成ArkTS模块 435 4361. 在 entry/src/main/cpp/types/libentry/Index.d.ts，定义ArkTS接口`runDemo()` 。内容如下： 437 438 ```ts 439 export const runDemo: (a: number[], b:Object) => Array<number>; 440 ``` 441 4422. 在 oh-package.json5 文件，将API与so相关联，成为一个完整的ArkTS模块： 443 444 ```json 445 { 446 "name": "libentry.so", 447 "types": "./Index.d.ts", 448 "version": "1.0.0", 449 "description": "MindSpore Lite inference module" 450 } 451 ``` 452 453### 调用封装的ArkTS模块进行推理并输出结果 454 455在 entry/src/main/ets/pages/Index.ets 中，调用封装的ArkTS模块，最后对推理结果进行处理。 456 457```ts 458// Index.ets 459import msliteNapi from 'libentry.so' 460 461@Entry 462@Component 463struct Index { 464 @State modelInputHeight: number = 224; 465 @State modelInputWidth: number = 224; 466 @State max: number = 0; 467 @State maxIndex: number = 0; 468 @State maxArray: Array<number> = []; 469 @State maxIndexArray: Array<number> = []; 470 471 build() { 472 Row() { 473 Column() { 474 Button() { 475 Text('photo') 476 .fontSize(30) 477 .fontWeight(FontWeight.Bold) 478 } 479 .type(ButtonType.Capsule) 480 .margin({ 481 top: 20 482 }) 483 .backgroundColor('#0D9FFB') 484 .width('40%') 485 .height('5%') 486 .onClick(() => { 487 488 let float32View = new Float32Array(this.modelInputHeight * this.modelInputWidth * 3); 489 // 图像输入和预处理。 490 // 调用c++的runDemo方法，完成图像输入和预处理后的buffer数据保存在float32View，具体可见上文图像输入和预处理中float32View的定义和处理。 491 console.info('MS_LITE_LOG: *** Start MSLite Demo ***'); 492 let output: Array<number> = msliteNapi.runDemo(Array.from(float32View), resMgr); 493 494 // 取分类占比的最大值 495 this.max = 0; 496 this.maxIndex = 0; 497 this.maxArray = []; 498 this.maxIndexArray = []; 499 let newArray = output.filter(value => value !== this.max); 500 for (let n = 0; n < 5; n++) { 501 this.max = output[0]; 502 this.maxIndex = 0; 503 for (let m = 0; m < newArray.length; m++) { 504 if (newArray[m] > this.max) { 505 this.max = newArray[m]; 506 this.maxIndex = m; 507 } 508 } 509 this.maxArray.push(Math.round(this.max * 10000)); 510 this.maxIndexArray.push(this.maxIndex); 511 // filter数组过滤函数 512 newArray = newArray.filter(value => value !== this.max); 513 } 514 console.info('MS_LITE_LOG: max:' + this.maxArray); 515 console.info('MS_LITE_LOG: maxIndex:' + this.maxIndexArray); 516 console.info('MS_LITE_LOG: *** Finished MSLite Demo ***'); 517 }) 518 }.width('100%') 519 } 520 .height('100%') 521 } 522} 523``` 524 525### 调测验证 526 5271. 在DevEco Studio中连接设备，点击Run entry，编译Hap，有如下显示： 528 529 ```shell 530 Launching com.samples.mindsporelitecdemo 531 $ hdc shell aa force-stop com.samples.mindsporelitecdemo 532 $ hdc shell mkdir data/local/tmp/xxx 533 $ hdc file send C:\Users\xxx\MindSporeLiteCDemo\entry\build\default\outputs\default\entry-default-signed.hap "data/local/tmp/xxx" 534 $ hdc shell bm install -p data/local/tmp/xxx 535 $ hdc shell rm -rf data/local/tmp/xxx 536 $ hdc shell aa start -a EntryAbility -b com.samples.mindsporelitecdemo 537 ``` 538 5392. 在设备屏幕点击photo按钮，选择图片，点击确定。设备屏幕显示所选图片的分类结果，在日志打印结果中，过滤关键字”MS_LITE“，可得到如下结果： 540 541 ```verilog 542 08-05 17:15:52.001 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: PhotoViewPicker.select successfully, photoSelectResult uri: {"photoUris":["file://media/Photo/13/IMG_1501955351_012/plant.jpg"]} 543 ... 544 08-05 17:15:52.627 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: crop info.width = 224 545 08-05 17:15:52.627 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: crop info.height = 224 546 08-05 17:15:52.628 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: Succeeded in reading image pixel data, buffer: 200704 547 08-05 17:15:52.971 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: float32View data: float32View data: 1.2385478019714355 1.308123230934143 1.4722440242767334 1.2385478019714355 1.308123230934143 1.4722440242767334 1.2385478019714355 1.308123230934143 1.4722440242767334 1.2385478019714355 1.308123230934143 1.4722440242767334 1.2385478019714355 1.308123230934143 1.4722440242767334 1.2385478019714355 1.308123230934143 1.4722440242767334 1.2385478019714355 1.308123230934143 548 08-05 17:15:52.971 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: *** Start MSLite Demo *** 549 08-05 17:15:53.454 4684-4684 A00000/[MSLiteNapi] pid-4684 I MS_LITE_LOG: Build MSLite model success. 550 08-05 17:15:53.753 4684-4684 A00000/[MSLiteNapi] pid-4684 I MS_LITE_LOG: Run MSLite model Predict success. 551 08-05 17:15:53.753 4684-4684 A00000/[MSLiteNapi] pid-4684 I MS_LITE_LOG: Get model outputs: 552 08-05 17:15:53.753 4684-4684 A00000/[MSLiteNapi] pid-4684 I MS_LITE_LOG: - Tensor 0 name is: Default/head-MobileNetV2Head/Sigmoid-op466. 553 08-05 17:15:53.753 4684-4684 A00000/[MSLiteNapi] pid-4684 I MS_LITE_LOG: - Tensor data is: 554 08-05 17:15:53.753 4684-4684 A00000/[MSLiteNapi] pid-4684 I MS_LITE_LOG: 3.43385e-06 1.40285e-05 9.11969e-07 4.91007e-05 9.50266e-07 3.94537e-07 0.0434676 3.97196e-05 0.00054832 0.000246202 1.576e-05 3.6494e-06 1.23553e-05 0.196977 5.3028e-05 3.29346e-05 4.90475e-07 1.66109e-06 7.03273e-06 8.83677e-07 3.1365e-06 555 08-05 17:15:53.781 4684-4684 A03d00/JSAPP pid-4684 W MS_LITE_WARN: output length = 500 ;value = 0.0000034338463592575863,0.000014028532859811094,9.119685273617506e-7,0.000049100715841632336,9.502661555416125e-7,3.945370394831116e-7,0.04346757382154465,0.00003971960904891603,0.0005483203567564487,0.00024620210751891136,0.000015759984307806008,0.0000036493988773145247,0.00001235533181898063,0.1969769448041916,0.000053027983085485175,0.000032934600312728435,4.904751449430478e-7,0.0000016610861166554969,0.000007032729172351537,8.836767619868624e-7 556 08-05 17:15:53.831 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: max:9497,7756,1970,435,46 557 08-05 17:15:53.831 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: maxIndex:323,46,13,6,349 558 08-05 17:15:53.831 4684-4684 A03d00/JSAPP pid-4684 I MS_LITE_LOG: *** Finished MSLite Demo *** 559 ``` 560 561 562### 效果示意 563 564在设备上，点击photo按钮，选择相册中的一张图片，点击确定。在图片下方显示此图片占比前4的分类信息。 565 566  567 568  569 570## 相关实例 571 572针对使用MindSpore Lite进行图像分类应用的开发，有以下相关实例可供参考： 573 574- [基于Native接口的MindSpore Lite应用开发（C/C++）（API11）](https://gitcode.com/openharmony/applications_app_samples/tree/master/code/DocsSample/ApplicationModels/MindSporeLiteCDemo) 575 576
