ANDROİD’DE MNIST VERİ SETİYLE ÖNCEDEN EĞİTİLMİŞ TENSORFLOW MODELİNİ KULLANMA | TENSORFLOW | ANDROİD -- http://www.bulentsiyah.com/androidde-mnist-veri-setiyle-onceden-egitilmis-tensorflow-modelini-kullanma-tensorflow-android/

Advanced_MNIST.py

import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

mnist_data = input_data.read_data_sets('MNIST_data', one_hot=True)


# Function to create a weight neuron using a random number. Training will assign a real weight later
def weight_variable(shape, name):
    initial = tf.truncated_normal(shape, stddev=0.1)
    return tf.Variable(initial, name=name)


# Function to create a bias neuron. Bias of 0.1 will help to prevent any 1 neuron from being chosen too often
def biases_variable(shape, name):
    initial = tf.constant(0.1, shape=shape)
    return tf.Variable(initial, name=name)


# Function to create a convolutional neuron. Convolutes input from 4d to 2d. This helps streamline inputs
def conv_2d(x, W, name):
    return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME', name=name)


# Function to create a neuron to represent the max input. Helps to make the best prediction for what comes next
def max_pool(x, name):
    return tf.nn.max_pool(x, ksize=[1, 2, 2, 1], strides=[1, 2, 2, 1], padding='SAME', name=name)


# A way to input images (as 784 element arrays of pixel values 0 - 1)
x_input = tf.placeholder(dtype=tf.float32, shape=[None, 784], name='x_input')
# A way to input labels to show model what the correct answer is during training
y_input = tf.placeholder(dtype=tf.float32, shape=[None, 10], name='y_input')

# First convolutional layer - reshape/resize images
# A weight variable that examines batches of 5x5 pixels, returns 32 features (1 feature per bit value in 32 bit float)
W_conv1 = weight_variable([5, 5, 1, 32], 'W_conv1')
# Bias variable to add to each of the 32 features
b_conv1 = biases_variable([32], 'b_conv1')
# Reshape each input image into a 28 x 28 x 1 pixel matrix
x_image = tf.reshape(x_input, [-1, 28, 28, 1], name='x_image')
# Flattens filter (W_conv1) to [5 * 5 * 1, 32], multiplies by [None, 28, 28, 1] to associate each 5x5 batch with the
# 32 features, and adds biases
h_conv1 = tf.nn.relu(conv_2d(x_image, W_conv1, name='conv1') + b_conv1, name='h_conv1')
# Takes windows of size 2x2 and computes a reduction on the output of h_conv1 (computes max, used for better prediction)
# Images are reduced to size 14 x 14 for analysis
h_pool1 = max_pool(h_conv1, name='h_pool1')

# Second convolutional layer, reshape/resize images
# Does mostly the same as above but converts each 32 unit output tensor from layer 1 to a 64 feature tensor
W_conv2 = weight_variable([5, 5, 32, 64], 'W_conv2')
b_conv2 = biases_variable([64], 'b_conv2')
h_conv2 = tf.nn.relu(conv_2d(h_pool1, W_conv2, name='conv2') + b_conv2, name='h_conv2')
# Images at this point are reduced to size 7 x 7 for analysis
h_pool2 = max_pool(h_conv2, name='h_pool2')

# First dense layer, performing calculation based on previous layer output
# Each image is 7 x 7 at the end of the previous section and outputs 64 features, we want 32 x 32 neurons = 1024
W_dense1 = weight_variable([7 * 7 * 64, 1024], name='W_dense1')
# bias variable added to each output feature
b_dense1 = biases_variable([1024], name='b_dense1')
# Flatten each of the images into size [None, 7 x 7 x 64]
h_pool_flat = tf.reshape(h_pool2, [-1, 7 * 7 * 64], name='h_pool_flat')
# Multiply weights by the outputs of the flatten neuron and add biases
h_dense1 = tf.nn.relu(tf.matmul(h_pool_flat, W_dense1, name='matmul_dense1') + b_dense1, name='h_dense1')

# Dropout layer prevents overfitting or recognizing patterns where none exist
# Depending on what value we enter into keep_prob, it will apply or not apply dropout layer
keep_prob = tf.placeholder(dtype=tf.float32, name='keep_prob')
# Dropout layer will be applied during training but not testing or predicting
h_drop1 = tf.nn.dropout(h_dense1, keep_prob, name='h_drop1')

# Readout layer used to format output
# Weight variable takes inputs from each of the 1024 neurons from before and outputs an array of 10 elements
W_readout1 = weight_variable([1024, 10], name='W_readout1')
# Apply bias to each of the 10 outputs
b_readout1 = biases_variable([10], name='b_readout1')
# Perform final calculation by multiplying each of the neurons from dropout layer by weights and adding biases
y_readout1 = tf.add(tf.matmul(h_drop1, W_readout1, name='matmul_readout1'), b_readout1, name='y_readout1')

# Softmax cross entropy loss function compares expected answers (labels) vs actual answers (logits)
cross_entropy_loss = tf.reduce_mean(tf.nn.softmax_cross_entropy_with_logits(labels=y_input, logits=y_readout1))
# Adam optimizer aims to minimize loss
train_step = tf.train.AdamOptimizer(0.0001).minimize(cross_entropy_loss)
# Compare actual vs expected outputs to see if highest number is at the same index, true if they match and false if not
correct_prediction = tf.equal(tf.argmax(y_input, 1), tf.argmax(y_readout1, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))

# Used to save the graph and weights
saver = tf.train.Saver()

# Run in with statement so session only exists within it
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())

    # Save the graph shape and node names to pbtxt file
    tf.train.write_graph(sess.graph_def, '.', 'advanced_mnist.pbtxt', False)

    # Train the model, running through data 20000 times in batches of 50
    # Print out step # and accuracy every 100 steps and final accuracy at the end of training
    # Train by running train_step and apply dropout by setting keep_prob to 0.5
    for i in range(20000):
        batch = mnist_data.train.next_batch(50)
        if i % 100 == 0:
            train_accuracy = accuracy.eval(feed_dict={x_input: batch[0], y_input: batch[1], keep_prob: 1.0})
            print("step %d, training accuracy %g" %(i, train_accuracy))
        train_step.run(feed_dict={x_input: batch[0], y_input: batch[1], keep_prob: 0.5})
    print("test accuracy %g" % accuracy.eval(feed_dict={x_input: mnist_data.test.images,
                                                        y_input: mnist_data.test.labels, keep_prob: 1.0}))

    # Save the session with graph shape and node weights
    saver.save(sess, 'advanced_mnist.ckpt')

    # Make a prediction
    print(sess.run(y_readout1, feed_dict={x_input: [mnist_data.test.images[0]], keep_prob: 1.0}))

build.gradle

apply plugin: 'com.android.application'

android {
    compileSdkVersion 27
    defaultConfig {
        applicationId "bulentsiyah.com.mlkit"
        minSdkVersion 22
        targetSdkVersion 27
        versionCode 1
        versionName "1.0"
        testInstrumentationRunner "android.support.test.runner.AndroidJUnitRunner"
    }
    buildTypes {
        release {
            minifyEnabled false
            proguardFiles getDefaultProguardFile('proguard-android.txt'), 'proguard-rules.pro'
        }
    }

    sourceSets {
        main {
            jniLibs.srcDirs = ['libs']
        }
    }
}

dependencies {
    implementation fileTree(dir: 'libs', include: ['*.jar'])
    implementation 'com.android.support:appcompat-v7:27.1.1'
    implementation 'com.android.support.constraint:constraint-layout:1.1.0'
    testImplementation 'junit:junit:4.12'
    androidTestImplementation 'com.android.support.test:runner:1.0.2'
    androidTestImplementation 'com.android.support.test.espresso:espresso-core:3.0.2'
}

Freeze_Graph.py

import tensorflow as tf
from tensorflow.python.tools import freeze_graph, optimize_for_inference_lib

# Saving the graph as a pb file taking data from pbtxt and ckpt files and providing a few operations
freeze_graph.freeze_graph('advanced_mnist.pbtxt',
                          '',
                          True,
                          'advanced_mnist.ckpt',
                          'y_readout1',
                          'save/restore_all',
                          'save/Const:0',
                          'frozen_advanced_mnist.pb',
                          True,
                          '')

# Read the data form the frozen graph pb file
input_graph_def = tf.GraphDef()
with tf.gfile.Open('frozen_advanced_mnist.pb', 'rb') as f:
    data = f.read()
    input_graph_def.ParseFromString(data)

# Optimize the graph with input and output nodes
output_graph_def = optimize_for_inference_lib.optimize_for_inference(
    input_graph_def,
    ['x_input', 'keep_prob'],
    ['y_readout1'],
    tf.float32.as_datatype_enum)

# Save the optimized graph to the optimized pb file
f = tf.gfile.FastGFile('optimized_advanced_mnist.pb', 'w')
f.write(output_graph_def.SerializeToString())

layout.xml

<?xml version="1.0" encoding="utf-8"?>
<android.support.constraint.ConstraintLayout
    xmlns:android="http://schemas.android.com/apk/res/android"
    xmlns:app="http://schemas.android.com/apk/res-auto"
    xmlns:tools="http://schemas.android.com/tools"
    android:layout_width="match_parent"
    android:layout_height="match_parent">

    <ImageView
        android:id="@+id/image_view"
        android:layout_width="300dp"
        android:layout_height="300dp"
        app:layout_constraintBottom_toTopOf="@+id/predict_button"
        app:layout_constraintLeft_toLeftOf="parent"
        app:layout_constraintRight_toRightOf="parent"
        app:layout_constraintTop_toTopOf="parent"
        android:contentDescription="@string/image_view_content_description"/>

    <Button
        android:id="@+id/predict_button"
        android:layout_width="wrap_content"
        android:layout_height="wrap_content"
        app:layout_constraintBottom_toTopOf="@+id/next_image_button"
        app:layout_constraintLeft_toLeftOf="parent"
        app:layout_constraintRight_toRightOf="parent"
        app:layout_constraintTop_toBottomOf="@+id/image_view"
        android:text="@string/predict_button_text"
        android:onClick="predictDigitClick"/>

    <Button
        android:id="@+id/next_image_button"
        android:layout_width="wrap_content"
        android:layout_height="wrap_content"
        app:layout_constraintBottom_toTopOf="@+id/text_view"
        app:layout_constraintLeft_toLeftOf="parent"
        app:layout_constraintRight_toRightOf="parent"
        app:layout_constraintTop_toBottomOf="@+id/predict_button"
        android:text="@string/next_image_button_text"
        android:onClick="loadNextImageClick"/>

    <TextView
        android:id="@+id/text_view"
        android:layout_width="wrap_content"
        android:layout_height="wrap_content"
        app:layout_constraintBottom_toBottomOf="parent"
        app:layout_constraintLeft_toLeftOf="parent"
        app:layout_constraintRight_toRightOf="parent"
        app:layout_constraintTop_toBottomOf="@+id/next_image_button"
        android:text="@string/text_view_text"
        android:textSize="20sp"/>

</android.support.constraint.ConstraintLayout>

MNISTOrnegi.java

package bulentsiyah.com.mlkit;

import android.graphics.Bitmap;
import android.graphics.BitmapFactory;
import android.support.v7.app.AppCompatActivity;
import android.os.Bundle;
import android.view.View;
import android.widget.ImageView;
import android.widget.TextView;
import org.tensorflow.contrib.android.TensorFlowInferenceInterface;

public class MNISTOrnegi extends AppCompatActivity {

    ImageView imageView;
    TextView textView;
    Boolean SimpleTrue_AdvancedFalse = false;

    static {
        System.loadLibrary("tensorflow_inference");
    }
    private static final String MODEL_FILE_Simple = "file:///android_asset/optimized_frozen_mnist_model.pb";
    private static final String INPUT_NODE = "x_input";
    private static final int[] INPUT_SHAPE = {1, 784};
    private static final String OUTPUT_NODE_Simple = "y_actual";

    private static final String MODEL_FILE = "file:///android_asset/optimized_advanced_mnist.pb";
    private static final int[] INPUT_SIZE = {1, 784};
    private static final String KEEP_PROB = "keep_prob";
    private static final int[] KEEP_PROB_SIZE = {1};
    private static final String OUTPUT_NODE = "y_readout1";

    private TensorFlowInferenceInterface inferenceInterface;

    private int imageListIndex = 9;
    private final int[] imageIDList = {
            R.drawable.digit0,
            R.drawable.digit1,
            R.drawable.digit2,
            R.drawable.digit3,
            R.drawable.digit4,
            R.drawable.digit5,
            R.drawable.digit6,
            R.drawable.digit7,
            R.drawable.digit8,
            R.drawable.digit9
    };

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

        imageView = (ImageView) findViewById(R.id.image_view);
        textView = (TextView) findViewById(R.id.text_view);

        inferenceInterface = new TensorFlowInferenceInterface();
        if(SimpleTrue_AdvancedFalse) {
            inferenceInterface.initializeTensorFlow(getAssets(), MODEL_FILE_Simple);
        }else{
            inferenceInterface.initializeTensorFlow(getAssets(), MODEL_FILE);
        }

    }

    public void predictDigitClick(View view) {
        float[] pixelBuffer = convertImage();
        float[] results = formPrediction(pixelBuffer);
        printResults(results);
    }
    private void printResults(float[] results) {
        float max = 0;
        float secondMax = 0;
        int maxIndex = 0;
        int secondMaxIndex = 0;
        for(int i = 0; i < 10; i++) {
            if (results[i] > max) {
                secondMax = max;
                secondMaxIndex = maxIndex;
                max = results[i];
                maxIndex = i;
            } else if (results[i] < max && results[i] > secondMax) {
                secondMax = results[i];
                secondMaxIndex = i;
            }
        }
        String output = "Model ilk tahmin: " + String.valueOf(maxIndex) +
                ", ikinci tahmin: " + String.valueOf(secondMaxIndex);
        textView.setText(output);
    }

    private float[] formPrediction(float[] pixelBuffer) {
        if(SimpleTrue_AdvancedFalse) {
            inferenceInterface.fillNodeFloat(INPUT_NODE, INPUT_SHAPE, pixelBuffer);
            inferenceInterface.runInference(new String[]{OUTPUT_NODE_Simple});
            float[] results = {0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
            inferenceInterface.readNodeFloat(OUTPUT_NODE_Simple, results);
            return results;
        }else{
            inferenceInterface.fillNodeFloat(INPUT_NODE, INPUT_SIZE, pixelBuffer);
            inferenceInterface.fillNodeFloat(KEEP_PROB, KEEP_PROB_SIZE, new float[] {0.5f});
            inferenceInterface.runInference(new String[] {OUTPUT_NODE});
            float[] outputs = new float[10];
            inferenceInterface.readNodeFloat(OUTPUT_NODE, outputs);
            return outputs;
        }
    }
    private float[] convertImage() {
        Bitmap imageBitmap = BitmapFactory.decodeResource(getResources(),
                imageIDList[imageListIndex]);
        //ölçekle
        imageBitmap = Bitmap.createScaledBitmap(imageBitmap, 28, 28, true);
        imageView.setImageBitmap(imageBitmap);
        int[] imageAsIntArray = new int[784];
        float[] imageAsFloatArray = new float[784];
        //getPixel metodu aldığı piksel kordinat parametrelerindeki pixel bilgilerine ilişkin int bir değer döndürür.
        // Bu değer android.graphics.Color sınıfına ilişkin int türden değerdir.
        // Pixel değerlerine ilişkin Alpha, Red, Green, Blue değerleri sırasıyla
        // Alpha, Red, Green, Blue static metotları ile elde edilebilir.
        imageBitmap.getPixels(imageAsIntArray, 0, 28, 0, 0, 28, 28);
        for (int i = 0; i < 784; i++) {
            imageAsFloatArray[i] = imageAsIntArray[i] / -16777216; //256x256x256= 16,777,216
        }
        return imageAsFloatArray;
    }
    public void loadNextImageClick(View view) {
        if (imageListIndex >= 9) {
            imageListIndex = 0;
        } else {
            imageListIndex += 1;
        }
        imageView.setImageDrawable(getDrawable(imageIDList[imageListIndex]));
    }
}

Simple_MNIST_Pycharm--Freeze_Graph.py

Freeze_Graph.py
import tensorflow as tf
from tensorflow.python.tools import freeze_graph, optimize_for_inference_lib

freeze_graph.freeze_graph(input_graph='mnist_model_n.pbtxt',
                          input_saver='',
                          input_binary=True,
                          input_checkpoint='mnist_model_n.ckpt',
                          output_node_names='y_actual',
                          restore_op_name='save/restore_all',
                          filename_tensor_name='save/Const:0',
                          output_graph='frozen_mnist_model.pb',
                          clear_devices=True,
                          initializer_nodes='')

input_graph_def = tf.GraphDef()
with tf.gfile.Open('frozen_mnist_model.pb', 'rb') as f:
    data = f.read()
    input_graph_def.ParseFromString(data)

output_graph_def = optimize_for_inference_lib.optimize_for_inference(input_graph_def=input_graph_def,
                                                                     input_node_names=['x_input'],
                                                                     output_node_names=['y_actual'],
                                                                     placeholder_type_enum=tf.float32.as_datatype_enum)

f = tf.gfile.FastGFile(name='optimized_frozen_mnist_model.pb',
                       mode='w')
f.write(file_content=output_graph_def.SerializeToString())

Simple_MNIST_Pycharm--MNIST_Model.py

MNIST_Model.py
import tensorflow as tf
from tensorflow.examples.tutorials.mnist import input_data

# Contains all of the images and labels (train and test) in the MNIST_data data set
mnist_data = input_data.read_data_sets('MNIST_data', one_hot=True)

#train_image = mnist_data.train.images[0]
#train_label = mnist_data.train.labels[0]
#print(train_image)
#print(train_label)

# y = Wx + b
# Input to the graph, takes in any number of images (784 element pixel arrays)
x_input = tf.placeholder(dtype=tf.float32, shape=[None, 784], name='x_input')
# Weights to be multiplied by input
W = tf.Variable(initial_value=tf.zeros(shape=[784, 10]), name='W')
# Biases to be added to weights * inputs
b = tf.Variable(initial_value=tf.zeros(shape=[10]), name='b')
# Actual model prediction based on input and current values of W and b
y_actual = tf.add(x=tf.matmul(a=x_input, b=W, name='matmul'), y=b, name='y_actual')
# Input to enter correct answer for comparison during training
y_expected = tf.placeholder(dtype=tf.float32, shape=[None, 10], name='y_expected')

# Cross entropy loss function because output is a list of possibilities (% certainty of the correct answer)
cross_entropy_loss = tf.reduce_mean(
    input_tensor=tf.nn.softmax_cross_entropy_with_logits(labels=y_expected, logits=y_actual),
    name='cross_entropy_loss')
# Classic gradient descent optimizer aims to minimize the difference between expected and actual values (loss)
optimizer = tf.train.GradientDescentOptimizer(learning_rate=0.5, name='optimizer')
train_step = optimizer.minimize(loss=cross_entropy_loss, name='train_step')

saver = tf.train.Saver()

# Create the session to run the nodes
session = tf.InteractiveSession()
session.run(tf.global_variables_initializer())

tf.train.write_graph(graph_or_graph_def=session.graph_def,
                     logdir='.',
                     name='mnist_model_n.pbtxt',
                     as_text=False)

# Train the model by fetching batches of 100 images and labels at a time and running train_step
# Run through the batches 1000 times (epochs)
for _ in range(1000):
    batch = mnist_data.train.next_batch(100)
    train_step.run(feed_dict={x_input: batch[0], y_expected: batch[1]})

saver.save(sess=session,save_path='./mnist_model_n.ckpt')

# Measure accuracy by comparing the predicted values to the correct values and calculating how many of them match
correct_prediction = tf.equal(x=tf.argmax(y_actual, 1), y=tf.argmax(y_expected, 1))
accuracy = tf.reduce_mean(tf.cast(x=correct_prediction, dtype=tf.float32))
print(accuracy.eval(feed_dict={x_input: mnist_data.test.images, y_expected: mnist_data.test.labels}))

# Test a prediction on a single image
print(session.run(fetches=y_actual, feed_dict={x_input: [mnist_data.test.images[0]]}))