android + tflite 分类APP开发-2

APP开发

build.gradle导入库

//implementation 'org.tensorflow:tensorflow-android:+'

implementation 'org.tensorflow:tensorflow-lite:2.4.0'

implementation 'org.tensorflow:tensorflow-lite-support:0.3.1

implementation 'org.tensorflow:tensorflow-lite-metadata:0.3.1'

加载模型

try {

tfLiteClassificationUtil = new TFLiteClassificationUtil(CONST.downPath + "/zjym.tflite");

Toast.makeText(MainActTflite.this, "模型加载成功!", Toast.LENGTH_SHORT).show();

} catch (Exception e) {

Toast.makeText(MainActTflite.this, "模型加载失败!", Toast.LENGTH_SHORT).show();

e.printStackTrace();

finish();

}

模型一般在assets目录下,在编译时会集成到APP中,不利于模型的迭代,这里模型保存在内部存储目录下。

分类预测

try { // 预测图像

FileInputStream fis = new FileInputStream(image_path);

imageView.setImageBitmap(BitmapFactory.decodeStream(fis));

long start = System.currentTimeMillis();

int[][] res2Arr = tfLiteClassificationUtil.predictImage(image_path);

long end = System.currentTimeMillis();

String show_text = "预测结果标签:" + (int) res2Arr[res2Arr.length-1][0] +

"\n名称:" + classNames.get((int) res2Arr[res2Arr.length-1][0]) +"概率:" + (float) res2Arr[res2Arr.length - 1][1] / 256 +

"\n名称:" + classNames.get((int) res2Arr[res2Arr.length-2][0]) +"概率:" + (float) res2Arr[res2Arr.length - 2][1] / 256 +

"\n名称:" + classNames.get((int) res2Arr[res2Arr.length-3][0]) +"概率:" + (float) res2Arr[res2Arr.length - 3][1] / 256 +

"\n时间:" + (end - start) + "ms";

textView.setText(show_text);

} catch (Exception e) {

e.printStackTrace();

}

res2Arr[res2Arr.length - 1][1] / 256,两个整数相除显示为0,添加(float)显示字符串

TFLiteClassificationUtil类功能模块

public TFLiteClassificationUtil(String modelPath) throws Exception {

File file = new File(modelPath);

if (!file.exists()) {

throw new Exception("model file is not exists!");

}

try {

Interpreter.Options options = new Interpreter.Options();

options.setNumThreads(NUM_THREADS);// 使用多线程预测

NnApiDelegate delegate = new NnApiDelegate();// 使用Android自带的API或者GPU加速

// GpuDelegate delegate = new GpuDelegate();

options.addDelegate(delegate);

tflite = new Interpreter(file, options);

// 获取输入,shape为{1, height, width, 3}

int[] imageShape = tflite.getInputTensor(tflite.getInputIndex("input_1")).shape();

DataType imageDataType = tflite.getInputTensor(tflite.getInputIndex("input_1")).dataType();

inputImageBuffer = new TensorImage(imageDataType);

// 获取输入,shape为{1, NUM_CLASSES}

int[] probabilityShape = tflite.getOutputTensor(tflite.getOutputIndex("Identity")).shape();

DataType probabilityDataType = tflite.getOutputTensor(tflite.getOutputIndex("Identity")).dataType();

//outputProbabilityBuffer = TensorBuffer.createFixedSize(probabilityShape, probabilityDataType);

outputProbabilityBuffer = TensorBuffer.createFixedSize(tflite.getOutputTensor(0).shape(), DataType.UINT8);

// 添加图像预处理方式

imageProcessor = new ImageProcessor.Builder()

.add(new ResizeOp(224, 224, ResizeOp.ResizeMethod.NEAREST_NEIGHBOR))

.add(new NormalizeOp(new float[] {0.0f}, new float[] {255.0f}))

.add(new QuantizeOp(0f, 0.003921569f))

.add(new CastOp(DataType.UINT8))

.build();

TensorProcessor probabilityPostProcessor = new TensorProcessor.Builder()

.add(new DequantizeOp((float) 0, (float) 0.00390625))

.add(new NormalizeOp(new float[]{0.0f}, new float[]{1.0f}))

.build();

} catch (Exception e) {

e.printStackTrace();

throw new Exception("load model fail!");

}

}

public int[][] predictImage(String image_path) throws Exception {

if (!new File(image_path).exists()) {

throw new Exception("image file is not exists!");

}

FileInputStream fis = new FileInputStream(image_path);

Bitmap bitmap = BitmapFactory.decodeStream(fis);

int[][] result = predictImage(bitmap);

if (bitmap.isRecycled()) {

bitmap.recycle();

}

return result;

}

// 重载方法,直接使用Bitmap预测

public int[][] predictImage(Bitmap bitmap) throws Exception {

return predict(bitmap);

}

private int[][] predict(Bitmap bmp) throws Exception {

inputImageBuffer = loadImage(bmp);

try {

tflite.run(inputImageBuffer.getBuffer(), outputProbabilityBuffer.getBuffer().rewind());

} catch (Exception e) {

throw new Exception("predict image fail! log:" + e);

}

int[] results = outputProbabilityBuffer.getIntArray();

Log.d("results", Arrays.toString(results));

int[][] arr = new int[results.length][2];

for (int i=0;i<results.length;i++) {

arr[i][0] = i;

arr[i][1] = results[i];

}

Arrays.sort(arr, Comparator.comparingInt(e -> e[1]));

//int l = getMaxResult(results);

return arr;//new float[]{l, results[l]};

}

tflite默认保存格式为UINT8,如果不加add(new CastOp(DataType.UINT8))可能显示

Cannot copy to a TensorFlowLite tensor (input_1) with 150528 bytes from a Java Buffer with 602112 bytes

默认的预训练模型是 EfficientNet-Lite0,如果为其他模型,其输入参数等也要修改。可通过下述方法查看。

Android Studio ->File ->open ->other ->tflite,打开tflite模型,build ->Make Project 会自动生成模型接口类,并移动模型到ml目录,查看类中模型参数。

相关推荐
Cao_Shixin攻城狮3 小时前
Flutter运行Android项目时显示java版本不兼容(Unsupported class file major version 65)的处理
android·java·flutter
呼啦啦呼啦啦啦啦啦啦6 小时前
利用pdfjs实现的pdf预览简单demo(包含翻页功能)
android·javascript·pdf
idjl8 小时前
Mysql测试题
android·adb
游戏开发爱好者810 小时前
iOS App 电池消耗管理与优化 提升用户体验的完整指南
android·ios·小程序·https·uni-app·iphone·webview
人生游戏牛马NPC1号11 小时前
学习 Flutter (四):玩安卓项目实战 - 中
android·学习·flutter
星辰也为你祝福h12 小时前
Android原生Dialog
android
wh_xia_jun12 小时前
基础分类模型及回归简介(一)
分类·数据挖掘·回归
梁同学与Android13 小时前
Android ---【CPU优化】需要优化的原因及优化的地方
android
Misha韩13 小时前
React Native 基础tabBar和自定义tabBar - bottom-tabs
android·react native
iHero14 小时前
【Nextcloud】在 Ubuntu 22.04.3 LTS 上的 Nextcloud Hub 10 (31.0.2) 后台任务cron 的优化
android·linux·ubuntu·nextcloud