开发者社区> 问答> 正文

Android分类应用程序与Tensorflow Lite模型崩溃

我与黑色素瘤细调CNN Mobilenet癌症数据集产生黑色素瘤和Non-Melanoma类分类模型,然后我转换Tensorflow Lite模式,但是当我开发它在移动,然后单击“分类”按钮程序崩溃告诉我不幸InceptionTutorial已经停止”。我的python调优代码如下:https://www.kaggle.com/gabrielmv/melanoma-classifie-mobilenet/notebook

from keras.models import Model
from keras.layers import Dense, Dropout
from keras.preprocessing.image import ImageDataGenerator
from keras.optimizers import Adam
from keras.callbacks import ReduceLROnPlateau, ModelCheckpoint, EarlyStopping
from keras.applications.mobilenet import MobileNet, preprocess_input
import numpy as np
import itertools
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt


def plot_confusion_matrix(cm, classes, normalize=False, title='Confusion matrix', cmap=plt.cm.Blues):
    if normalize:
        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]
        print("Normalized confusion matrix")
    else:
        print('Confusion matrix, without normalization')

    print(cm)

    plt.imshow(cm, interpolation='nearest', cmap=cmap)
    plt.title(title)
    plt.colorbar()
    tick_marks = np.arange(len(classes))
    plt.xticks(tick_marks, classes, rotation=45)
    plt.yticks(tick_marks, classes)

    fmt = '.2f' if normalize else 'd'
    thresh = cm.max() / 2.
    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):
        plt.text(j, i, format(cm[i, j], fmt),
                 horizontalalignment="center",
                 color="white" if cm[i, j] > thresh else "black")

    plt.ylabel('True label')
    plt.xlabel('Predicted label')
    plt.tight_layout()

def plot_training_curves(history):
    acc = history.history['acc']
    val_acc = history.history['val_acc']
    loss = history.history['loss']
    val_loss = history.history['val_loss']

    epochs = range(1, len(acc) + 1)

    plt.plot(epochs, loss, 'r', label='Training loss')
    plt.plot(epochs, val_loss, 'g', label='Validation loss')
    plt.title('Training and validation loss')
    plt.legend()
    plt.figure()

    plt.plot(epochs, acc, 'r', label='Training acc')
    plt.plot(epochs, val_acc, 'g', label='Validation acc')
    plt.title('Training and validation accuracy')
    plt.legend()
    plt.figure()

    plt.show()

def print_results(cm):
    tp = cm[0, 0]
    tn = cm[1, 1]
    fn = cm[0, 1]
    fp = cm[1, 0]

    accuracy = ((tp + tn) / (tp + tn + fp + fn)) * 100

    sensitivity = (tp / (tp + fn)) * 100

    specificity = (tn / ( tn + fp )) * 100

    print ('Accuracy: ',  accuracy)

    print ('Sensitivity: ', sensitivity)

    print ('Specificity: ',  specificity)

def fine_tune_mobile_net(train_batches, train_steps, class_weights, valid_batches, val_steps, file_path):
    mobile = MobileNet() # mobile.summary()

    x = mobile.layers[-6].output
    x = Dropout(0.5)(x)
    predictions = Dense(2, activation='softmax')(x)    

    model = Model(inputs = mobile.input, outputs = predictions)

    for layer in mobile.layers:
        layer.trainable = False

    model.compile(Adam(lr = 0.003), 
                  loss = 'categorical_crossentropy', 
                  metrics = ['accuracy'])

    model.fit_generator(train_batches,
                          steps_per_epoch = train_steps,
                          class_weight = class_weights,
                          validation_data = valid_batches,
                          validation_steps = val_steps,
                          epochs = 50,
                          verbose = 1,
                          callbacks = callbacks)


    model.load_weights(file_path)
    print ('*** Fine Tunning MobileNet ***')

    for layer in model.layers[:-23]:
        layer.trainable = False

    model.compile(Adam(lr = 0.003), 
                  loss = 'categorical_crossentropy', 
                  metrics = ['accuracy'])

    history = model.fit_generator(train_batches,
                          steps_per_epoch = train_steps,
                          class_weight = class_weights,
                          validation_data = valid_batches,
                          validation_steps = val_steps,
                          epochs = 75,
                          verbose = 1,
                          callbacks = callbacks)

    return model, history

def save_model(model, file_path):
    model_json = model.to_json()
    with open('model.json', 'w') as json_file:
        json_file.write(model_json)


file_path = 'weights-mobilenet-2.0.h5'

callbacks = [
        ModelCheckpoint(file_path, monitor = 'val_acc', verbose = 1, save_best_only = True, mode = 'max'),
        ReduceLROnPlateau(monitor = 'val_loss', factor = 0.2, patience = 8, verbose = 1, mode = 'min', min_lr = 0.00001),
        EarlyStopping(monitor = 'val_loss', min_delta = 1e-10, patience = 15, verbose = 1)
        ]

training_path = '../input/dermmel/DermMel/train_sep'
validation_path = '../input/dermmel/DermMel/valid'
test_path = '../input/dermmel/DermMel/test'

num_train_samples = 10682
num_val_samples = 3562
num_test_samples = 3561

train_batch_size = 16
val_batch_size = 16
test_batch_size = 16

train_steps = np.ceil(num_train_samples / train_batch_size)
val_steps = np.ceil(num_val_samples / val_batch_size)
test_steps = np.ceil(num_val_samples / val_batch_size)

class_weights = {
        0: 4.1, # melanoma
        1: 1.0 # non-melanoma
}

train_batches = ImageDataGenerator(rescale = 1./255).flow_from_directory(training_path,
                                    target_size = (224, 224),
                                    batch_size = val_batch_size,
                                    class_mode = 'categorical')

valid_batches = ImageDataGenerator(rescale = 1./255).flow_from_directory(validation_path,
                                        target_size = (224, 224),
                                        batch_size = val_batch_size,
                                        class_mode = 'categorical')

test_batches = ImageDataGenerator(rescale = 1./255).flow_from_directory(test_path,
                                          target_size = (224, 224),
                                          batch_size = test_batch_size,
                                          class_mode = 'categorical',
                                          shuffle = False)


model, history = fine_tune_mobile_net(train_batches, train_steps, class_weights, valid_batches, val_steps, file_path)


save_model(model, file_path)
model.load_weights(file_path)

test_labels = test_batches.classes

predictions = model.predict_generator(test_batches, steps = val_steps, verbose = 1)

cm = confusion_matrix(test_labels, predictions.argmax(axis=1))

然后使用之前用于多对象分类的Android源代码,从(https://github.com/soum-io/TensorFlowLiteInceptionTutorial)中定制我的黑色素瘤分类应用程序(https://github.com/soum-io/TensorFlowLiteInceptionTutorial)进行两个主要活动 ChooseModel.java

package com.soumio.inceptiontutorial;

import android.Manifest;
import android.content.ContentValues;
import android.content.Intent;
import android.content.pm.ActivityInfo;
import android.content.pm.PackageManager;
import android.net.Uri;
import android.os.Build;
import android.provider.MediaStore;
import android.support.annotation.NonNull;
import android.support.v4.app.ActivityCompat;
import android.support.v4.content.ContextCompat;
import android.support.v7.app.AppCompatActivity;
import android.os.Bundle;
import android.util.Log;
import android.view.View;
import android.widget.Button;
import android.widget.Toast;

import com.soundcloud.android.crop.Crop;

import java.io.File;

public class ChooseModel extends AppCompatActivity {

    // button for each available classifier
    private Button inceptionFloat;
    private Button inceptionQuant;

    // for permission requests
    public static final int REQUEST_PERMISSION = 300;

    // request code for permission requests to the os for image
    public static final int REQUEST_IMAGE = 100;

    // will hold uri of image obtained from camera
    private Uri imageUri;

    // string to send to next activity that describes the chosen classifier
    private String chosen;

    //boolean value dictating if chosen model is quantized version or not.
    private boolean quant;


    @Override
    protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        setContentView(R.layout.activity_choose_model);

        // request permission to use the camera on the user's phone
        if (ActivityCompat.checkSelfPermission(this.getApplicationContext(), android.Manifest.permission.CAMERA) != PackageManager.PERMISSION_GRANTED){
            ActivityCompat.requestPermissions(this, new String[] {android.Manifest.permission.CAMERA}, REQUEST_PERMISSION);
        }

        // request permission to write data (aka images) to the user's external storage of their phone
        if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.M
                && ContextCompat.checkSelfPermission(this, Manifest.permission.WRITE_EXTERNAL_STORAGE) != PackageManager.PERMISSION_GRANTED) {
            ActivityCompat.requestPermissions(this, new String[]{Manifest.permission.WRITE_EXTERNAL_STORAGE},
                    REQUEST_PERMISSION);
        }

        // request permission to read data (aka images) from the user's external storage of their phone
        if (Build.VERSION.SDK_INT >= Build.VERSION_CODES.M
                && ContextCompat.checkSelfPermission(this, Manifest.permission.READ_EXTERNAL_STORAGE) != PackageManager.PERMISSION_GRANTED) {
            ActivityCompat.requestPermissions(this, new String[]{Manifest.permission.READ_EXTERNAL_STORAGE},
                    REQUEST_PERMISSION);
        }

        // on click for inception float model
        inceptionFloat = (Button)findViewById(R.id.inception_float);
        inceptionFloat.setOnClickListener(new View.OnClickListener() {
            @Override
            public void onClick(View view) {
                // filename in assets
                chosen = "inception_float.tflite";
                // model in not quantized
                quant = false;
                // open camera
                openCameraIntent();
            }
        });

        // on click for inception quant model
        inceptionQuant = (Button)findViewById(R.id.inception_quant);
        inceptionQuant.setOnClickListener(new View.OnClickListener() {
            @Override
            public void onClick(View view) {
                // filename in assets
                chosen = "inception_quant.tflite";
                // model in not quantized
                quant = true;
                // open camera
                openCameraIntent();
            }
        });
    }

    // opens camera for user
    private void openCameraIntent(){
        ContentValues values = new ContentValues();
        values.put(MediaStore.Images.Media.TITLE, "New Picture");
        values.put(MediaStore.Images.Media.DESCRIPTION, "From your Camera");
        // tell camera where to store the resulting picture
        imageUri = getContentResolver().insert(
                MediaStore.Images.Media.EXTERNAL_CONTENT_URI, values);
        Intent intent = new Intent(MediaStore.ACTION_IMAGE_CAPTURE);
        intent.putExtra(MediaStore.EXTRA_OUTPUT, imageUri);
        setRequestedOrientation(ActivityInfo.SCREEN_ORIENTATION_PORTRAIT);
        // start camera, and wait for it to finish
        startActivityForResult(intent, REQUEST_IMAGE);
    }

    // checks that the user has allowed all the required permission of read and write and camera. If not, notify the user and close the application
    @Override
    public void onRequestPermissionsResult(final int requestCode, @NonNull final String[] permissions, @NonNull final int[] grantResults) {
        super.onRequestPermissionsResult(requestCode, permissions, grantResults);
        if (requestCode == REQUEST_PERMISSION) {
            if (!(grantResults.length > 0 && grantResults[0] == PackageManager.PERMISSION_GRANTED)) {
                Toast.makeText(getApplicationContext(),"This application needs read, write, and camera permissions to run. Application now closing.",Toast.LENGTH_LONG);
                System.exit(0);
            }
        }
    }

    // dictates what to do after the user takes an image, selects and image, or crops an image
    @Override
    protected void onActivityResult(int requestCode, int resultCode, Intent data){
        super.onActivityResult(requestCode, resultCode, data);
        // if the camera activity is finished, obtained the uri, crop it to make it square, and send it to 'Classify' activity
        if(requestCode == REQUEST_IMAGE && resultCode == RESULT_OK) {
            try {
                Uri source_uri = imageUri;
                Uri dest_uri = Uri.fromFile(new File(getCacheDir(), "cropped"));
                // need to crop it to square image as CNN's always required square input
                Crop.of(source_uri, dest_uri).asSquare().start(ChooseModel.this);
            } catch (Exception e) {
                e.printStackTrace();
            }
        }

        // if cropping acitivty is finished, get the resulting cropped image uri and send it to 'Classify' activity
        else if(requestCode == Crop.REQUEST_CROP && resultCode == RESULT_OK){
            imageUri = Crop.getOutput(data);
            Intent i = new Intent(ChooseModel.this, Classify.class);
            // put image data in extras to send
            i.putExtra("resID_uri", imageUri);
            // put filename in extras
            i.putExtra("chosen", chosen);
            // put model type in extras
            i.putExtra("quant", quant);
            // send other required data
            startActivity(i);
        }
    }
}

Classify.java

package com.soumio.inceptiontutorial;

import android.content.Intent;
import android.content.res.AssetFileDescriptor;
import android.graphics.Bitmap;
import android.graphics.Matrix;
import android.graphics.drawable.BitmapDrawable;
import android.net.Uri;
import android.os.SystemClock;
import android.provider.MediaStore;
import android.support.v7.app.AppCompatActivity;
import android.os.Bundle;
import android.view.View;
import android.widget.Button;
import android.widget.ImageView;
import android.widget.TextView;

import org.tensorflow.lite.Interpreter;

import java.io.BufferedReader;
import java.io.FileInputStream;
import java.io.IOException;
import java.io.InputStreamReader;
import java.nio.ByteBuffer;
import java.nio.ByteOrder;
import java.nio.MappedByteBuffer;
import java.nio.channels.FileChannel;
import java.util.AbstractMap;
import java.util.ArrayList;
import java.util.Comparator;
import java.util.List;
import java.util.Map;
import java.util.PriorityQueue;

public class Classify extends AppCompatActivity {

    // presets for rgb conversion
    private static final int RESULTS_TO_SHOW = 3;
    private static final int IMAGE_MEAN = 128;
    private static final float IMAGE_STD = 128.0f;

    // options for model interpreter
    private final Interpreter.Options tfliteOptions = new Interpreter.Options();
    // tflite graph
    private Interpreter tflite;
    // holds all the possible labels for model
    private List<String> labelList;
    // holds the selected image data as bytes
    private ByteBuffer imgData = null;
    // holds the probabilities of each label for non-quantized graphs
    private float[][] labelProbArray = null;
    // holds the probabilities of each label for quantized graphs
    private byte[][] labelProbArrayB = null;
    // array that holds the labels with the highest probabilities
    private String[] topLables = null;
    // array that holds the highest probabilities
    private String[] topConfidence = null;


    // selected classifier information received from extras
    private String chosen;
    private boolean quant;

    // input image dimensions for the Inception Model
    private int DIM_IMG_SIZE_X = 244;
    private int DIM_IMG_SIZE_Y = 244;
    private int DIM_PIXEL_SIZE = 3;

    // int array to hold image data
    private int[] intValues;

    // activity elements
    private ImageView selected_image;
    private Button classify_button;
    private Button back_button;
    private TextView label1;
    private TextView label2;
    private TextView label3;
    private TextView Confidence1;
    private TextView Confidence2;
    private TextView Confidence3;

    // priority queue that will hold the top results from the CNN
    private PriorityQueue<Map.Entry<String, Float>> sortedLabels =
            new PriorityQueue<>(
                    RESULTS_TO_SHOW,
                    new Comparator<Map.Entry<String, Float>>() {
                        @Override
                        public int compare(Map.Entry<String, Float> o1, Map.Entry<String, Float> o2) {
                            return (o1.getValue()).compareTo(o2.getValue());
                        }
                    });

    @Override
    protected void onCreate(Bundle savedInstanceState) {
        // get all selected classifier data from classifiers
        chosen = (String) getIntent().getStringExtra("chosen");
        quant = (boolean) getIntent().getBooleanExtra("quant", false);

        // initialize array that holds image data
        intValues = new int[DIM_IMG_SIZE_X * DIM_IMG_SIZE_Y];

        super.onCreate(savedInstanceState);

        //initilize graph and labels
        try{
            tflite = new Interpreter(loadModelFile(), tfliteOptions);
            labelList = loadLabelList();
        } catch (Exception ex){
            ex.printStackTrace();
        }

        // initialize byte array. The size depends if the input data needs to be quantized or not
        if(quant){
            imgData =
                    ByteBuffer.allocateDirect(
                            DIM_IMG_SIZE_X * DIM_IMG_SIZE_Y * DIM_PIXEL_SIZE);
        } else {
            imgData =
                    ByteBuffer.allocateDirect(
                            4 * DIM_IMG_SIZE_X * DIM_IMG_SIZE_Y * DIM_PIXEL_SIZE);
        }
        imgData.order(ByteOrder.nativeOrder());

        // initialize probabilities array. The datatypes that array holds depends if the input data needs to be quantized or not
        if(quant){
            labelProbArrayB= new byte[1][labelList.size()];
        } else {
            labelProbArray = new float[1][labelList.size()];
        }

        setContentView(R.layout.activity_classify);

        // labels that hold top three results of CNN
        label1 = (TextView) findViewById(R.id.label1);
        label2 = (TextView) findViewById(R.id.label2);
        label3 = (TextView) findViewById(R.id.label3);
        // displays the probabilities of top labels
        Confidence1 = (TextView) findViewById(R.id.Confidence1);
        Confidence2 = (TextView) findViewById(R.id.Confidence2);
        Confidence3 = (TextView) findViewById(R.id.Confidence3);
        // initialize imageView that displays selected image to the user
        selected_image = (ImageView) findViewById(R.id.selected_image);

        // initialize array to hold top labels
        topLables = new String[RESULTS_TO_SHOW];
        // initialize array to hold top probabilities
        topConfidence = new String[RESULTS_TO_SHOW];

        // allows user to go back to activity to select a different image
        back_button = (Button)findViewById(R.id.back_button);
        back_button.setOnClickListener(new View.OnClickListener() {
            @Override
            public void onClick(View view) {
                Intent i = new Intent(Classify.this, ChooseModel.class);
                startActivity(i);
            }
        });

        // classify current dispalyed image
        classify_button = (Button)findViewById(R.id.classify_image);
        classify_button.setOnClickListener(new View.OnClickListener() {
            @Override
            public void onClick(View view) {
                // get current bitmap from imageView
                Bitmap bitmap_orig = ((BitmapDrawable)selected_image.getDrawable()).getBitmap();
                // resize the bitmap to the required input size to the CNN
                Bitmap bitmap = getResizedBitmap(bitmap_orig, DIM_IMG_SIZE_X, DIM_IMG_SIZE_Y);
                // convert bitmap to byte array
                convertBitmapToByteBuffer(bitmap);
                // pass byte data to the graph
                if(quant){
                    tflite.run(imgData, labelProbArrayB);
                } else {
                    tflite.run(imgData, labelProbArray);
                }
                // display the results
                printTopKLabels();
            }
        });

        // get image from previous activity to show in the imageView
        Uri uri = (Uri)getIntent().getParcelableExtra("resID_uri");
        try {
            Bitmap bitmap = MediaStore.Images.Media.getBitmap(getContentResolver(), uri);
            selected_image.setImageBitmap(bitmap);
            // not sure why this happens, but without this the image appears on its side
            selected_image.setRotation(selected_image.getRotation() + 90);
        } catch (IOException e) {
            e.printStackTrace();
        }
    }

    // loads tflite grapg from file
    private MappedByteBuffer loadModelFile() throws IOException {
        AssetFileDescriptor fileDescriptor = this.getAssets().openFd(chosen);
        FileInputStream inputStream = new FileInputStream(fileDescriptor.getFileDescriptor());
        FileChannel fileChannel = inputStream.getChannel();
        long startOffset = fileDescriptor.getStartOffset();
        long declaredLength = fileDescriptor.getDeclaredLength();
        return fileChannel.map(FileChannel.MapMode.READ_ONLY, startOffset, declaredLength);
    }

    // converts bitmap to byte array which is passed in the tflite graph
    private void convertBitmapToByteBuffer(Bitmap bitmap) {
        if (imgData == null) {
            return;
        }
        imgData.rewind();
        bitmap.getPixels(intValues, 0, bitmap.getWidth(), 0, 0, bitmap.getWidth(), bitmap.getHeight());
        // loop through all pixels
        int pixel = 0;
        for (int i = 0; i < DIM_IMG_SIZE_X; ++i) {
            for (int j = 0; j < DIM_IMG_SIZE_Y; ++j) {
                final int val = intValues[pixel++];
                // get rgb values from intValues where each int holds the rgb values for a pixel.
                // if quantized, convert each rgb value to a byte, otherwise to a float
                if(quant){
                    imgData.put((byte) ((val >> 16) & 0xFF));
                    imgData.put((byte) ((val >> 8) & 0xFF));
                    imgData.put((byte) (val & 0xFF));
                } else {
                    imgData.putFloat((((val >> 16) & 0xFF)-IMAGE_MEAN)/IMAGE_STD);
                    imgData.putFloat((((val >> 8) & 0xFF)-IMAGE_MEAN)/IMAGE_STD);
                    imgData.putFloat((((val) & 0xFF)-IMAGE_MEAN)/IMAGE_STD);
                }

            }
        }
    }

    // loads the labels from the label txt file in assets into a string array
    private List<String> loadLabelList() throws IOException {
        List<String> labelList = new ArrayList<String>();
        BufferedReader reader =
                new BufferedReader(new InputStreamReader(this.getAssets().open("labels.txt")));
        String line;
        while ((line = reader.readLine()) != null) {
            labelList.add(line);
        }
        reader.close();
        return labelList;
    }

    // print the top labels and respective confidences
    private void printTopKLabels() {
        // add all results to priority queue
        for (int i = 0; i < labelList.size(); ++i) {
            if(quant){
                sortedLabels.add(
                        new AbstractMap.SimpleEntry<>(labelList.get(i), (labelProbArrayB[0][i] & 0xff) / 255.0f));
            } else {
                sortedLabels.add(
                        new AbstractMap.SimpleEntry<>(labelList.get(i), labelProbArray[0][i]));
            }
            if (sortedLabels.size() > RESULTS_TO_SHOW) {
                sortedLabels.poll();
            }
        }

        // get top results from priority queue
        final int size = sortedLabels.size();
        for (int i = 0; i < size; ++i) {
            Map.Entry<String, Float> label = sortedLabels.poll();
            topLables[i] = label.getKey();
            topConfidence[i] = String.format("%.0f%%",label.getValue()*100);
        }

        // set the corresponding textviews with the results
        label1.setText("1. "+topLables[2]);
        label2.setText("2. "+topLables[1]);
      //  label3.setText("3. "+topLables[0]);
        Confidence1.setText(topConfidence[2]);
        Confidence2.setText(topConfidence[1]);
       // Confidence3.setText(topConfidence[0]);
    }


    // resizes bitmap to given dimensions
    public Bitmap getResizedBitmap(Bitmap bm, int newWidth, int newHeight) {
        int width = bm.getWidth();
        int height = bm.getHeight();
        float scaleWidth = ((float) newWidth) / width;
        float scaleHeight = ((float) newHeight) / height;
        Matrix matrix = new Matrix();
        matrix.postScale(scaleWidth, scaleHeight);
        Bitmap resizedBitmap = Bitmap.createBitmap(
                bm, 0, 0, width, height, matrix, false);
        return resizedBitmap;
    }
}

有谁能帮我解决这个问题,谢谢! 问题来源StackOverflow 地址:/questions/59465930/android-classification-app-crashes-with-tensorflow-lite-model

展开
收起
kun坤 2019-12-25 15:55:09 1375 0
1 条回答
写回答
取消 提交回答
问答排行榜
最热
最新

相关电子书

更多
58同城Android客户端Walle框架演进与实践之路 立即下载
Android组件化实现 立即下载
蚂蚁聚宝Android秒级编译——Freeline 立即下载