Android TensorFlow支持和TensorFlow Lite for Android是两种用于在Android设备上部署和运行TensorFlow模型的不同方式。
Android TensorFlow支持是TensorFlow官方提供的一个库,它允许开发者在Android设备上使用TensorFlow模型进行预测。它使用TensorFlow原生的Java API,并且可以直接加载和运行TensorFlow SavedModel,FrozenModel和GraphDef模型。
以下是一个使用Android TensorFlow支持进行图像分类的示例代码:
import android.content.res.AssetFileDescriptor;
import android.content.res.AssetManager;
import android.graphics.Bitmap;
import android.graphics.BitmapFactory;
import android.os.Bundle;
import android.support.v7.app.AppCompatActivity;
import android.util.Log;
import android.widget.ImageView;
import android.widget.TextView;
import org.tensorflow.contrib.android.TensorFlowInferenceInterface;
import java.io.IOException;
import java.io.InputStream;
public class MainActivity extends AppCompatActivity {
private static final String MODEL_FILE = "file:///android_asset/model.pb";
private static final String INPUT_NODE = "input";
private static final String OUTPUT_NODE = "output";
private static final int INPUT_SIZE = 224;
private static final int NUM_CLASSES = 1000;
private TensorFlowInferenceInterface inferenceInterface;
private Bitmap inputBitmap;
private ImageView imageView;
private TextView textView;
@Override
protected void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
setContentView(R.layout.activity_main);
imageView = findViewById(R.id.imageView);
textView = findViewById(R.id.textView);
inferenceInterface = new TensorFlowInferenceInterface(getAssets(), MODEL_FILE);
try {
inputBitmap = getBitmapFromAsset("input.jpg");
imageView.setImageBitmap(inputBitmap);
float[] result = classifyImage(inputBitmap);
String label = getLabel(result);
textView.setText("Class: " + label);
} catch (IOException e) {
e.printStackTrace();
}
}
private float[] classifyImage(Bitmap bitmap) {
float[] inputFloats = preprocessImage(bitmap);
inferenceInterface.feed(INPUT_NODE, inputFloats, 1, INPUT_SIZE, INPUT_SIZE, 3);
inferenceInterface.run(new String[]{OUTPUT_NODE});
float[] outputFloats = new float[NUM_CLASSES];
inferenceInterface.fetch(OUTPUT_NODE, outputFloats);
return outputFloats;
}
private float[] preprocessImage(Bitmap bitmap) {
Bitmap resizedBitmap = Bitmap.createScaledBitmap(bitmap, INPUT_SIZE, INPUT_SIZE, false);
int[] intValues = new int[INPUT_SIZE * INPUT_SIZE];
float[] floatValues = new float[INPUT_SIZE * INPUT_SIZE * 3];
resizedBitmap.getPixels(intValues, 0, resizedBitmap.getWidth(), 0, 0, resizedBitmap.getWidth(), resizedBitmap.getHeight());
for (int i = 0; i < intValues.length; ++i) {
final int val = intValues[i];
floatValues[i * 3 + 0] = ((val >> 16) & 0xFF) / 255.0f;
floatValues[i * 3 + 1] = ((val >> 8) & 0xFF) / 255.0f;
floatValues[i * 3 + 2] = (val & 0xFF) / 255.0f;
}
return floatValues;
}
private String getLabel(float[] result) {
// Load labels from file
String labelFile = "file:///android_asset/labels.txt";
String actualFilename = labelFile.split("file:///android_asset/")[1];
AssetManager assetManager = getAssets();
InputStream labelsInput;
String[] labels = new String[NUM_CLASSES];
try {
labelsInput = assetManager.open(actualFilename);
int bytesRead = labelsInput.read();
StringBuilder sb = new StringBuilder();
int i = 0;
while (bytesRead != -1) {
if ((char) bytesRead == '\n') {
labels[i] = sb.toString();
sb = new StringBuilder();
i++;
} else {
sb.append((char) bytesRead);
}
bytesRead = labelsInput.read();
}
labelsInput.close();
} catch (IOException e) {
e.printStackTrace();
}
int maxIndex = 0;
float maxValue = result[0];
for (int i = 1; i < result.length; i++) {
if (result[i] > maxValue) {
maxIndex = i;
maxValue = result[i];
}
}
return labels[maxIndex];
}
private Bitmap getBitmapFromAsset(String fileName) throws IOException {
AssetManager assetManager = getAssets();
InputStream inputStream = null;
try