我无法准确确定问题出在哪里，但我知道 TFLite 模型接收位图的方式存在问题，因为我尝试通过调用 Camera intent 以不同的方式发送它，并且它通过这样做按预期工作，但是我无法调用 Camera Intent，因为它会拉起 Camera 屏幕，而我的目标是不这样做。
此代码的两个版本都使用相同的 tflite 模型，所以我知道这不是模型问题。
工作版本：
picture.setOnClickListener(new View.OnClickListener() {
@Override
public void onClick(View view) {
Intent cameraIntent = new Intent(MediaStore.ACTION_IMAGE_CAPTURE);
startActivityForResult(cameraIntent, 1);
}
public void classifyImage(Bitmap image){
try {
Model model = Model.newInstance(getApplicationContext());
TensorBuffer inputFeature0 = TensorBuffer.createFixedSize(new int[]{1, 224, 224, 3}, DataType.FLOAT32);
ByteBuffer byteBuffer = ByteBuffer.allocateDirect(4 * imageSize * imageSize * 3);
byteBuffer.order(ByteOrder.nativeOrder());
int [] intValues = new int[imageSize * imageSize];
image.getPixels(intValues, 0, image.getWidth(), 0, 0, image.getWidth(), image.getHeight());
int pixel = 0;
for(int i = 0; i < imageSize; i++){
for(int j = 0; j < imageSize; j++){
int val = intValues[pixel++]; // RGB
byteBuffer.putFloat(((val >> 16) & 0xFF) * (1.f / 255.f));
byteBuffer.putFloat(((val >> 8) & 0xFF) * (1.f / 255.f));
byteBuffer.putFloat((val & 0xFF) * (1.f / 255.f));
}
}
inputFeature0.loadBuffer(byteBuffer);
Model.Outputs outputs = model.process(inputFeature0);
TensorBuffer outputFeature0 = outputs.getOutputFeature0AsTensorBuffer();
float[] confidences = outputFeature0.getFloatArray();
int maxPos = 0;
float maxConfidence = 0;
for(int i = 0; i < confidences.length; i++){
if(confidences[i] > maxConfidence){
maxConfidence = confidences[i];
maxPos = i;
}
}
String[] classes = {"Banana", "Orange", "Pen", "Sticky Notes"};
result.setText(classes[maxPos]);
String s = "";
for(int i = 0; i < classes.length; i++){
s += String.format("%s: %.1f%%\n", classes[i], confidences[i] * 100);
}
confidence.setText(s);
model.close();
}
@Override
public void onActivityResult(int requestCode, int resultCode, @Nullable Intent data) {
if (requestCode == 1 && resultCode == RESULT_OK) {
Bitmap image = (Bitmap) data.getExtras().get("data");
int dimension = Math.min(image.getWidth(), image.getHeight());
image = ThumbnailUtils.extractThumbnail(image, dimension, dimension);
imageView.setImageBitmap(image);
image = Bitmap.createScaledBitmap(image, imageSize, imageSize, false);
classifyImage(image);
}
super.onActivityResult(requestCode, resultCode, data);
}
}
非工作版本： 无论图像如何，此版本将始终返回相同的置信度（根据使用的不同 tflite 模型而变化）。
public class MainActivity extends AppCompatActivity implements ImageAnalysis.Analyzer, View.OnClickListener {
private ListenableFuture<ProcessCameraProvider> cameraProviderFuture;
PreviewView previewView;
private ImageCapture imageCapture;
private Button bCapture;
ImageView imageView;
@Override
protected void onCreate(Bundle savedInstanceState) {
cameraProviderFuture = ProcessCameraProvider.getInstance(this);
cameraProviderFuture.addListener(() -> {
try {
ProcessCameraProvider cameraProvider = cameraProviderFuture.get();
startCameraX(cameraProvider);
} catch (ExecutionException | InterruptedException e) {
e.printStackTrace();
}
}, getExecutor());
}
Executor getExecutor() {
return ContextCompat.getMainExecutor(this);
}
private void startCameraX(ProcessCameraProvider cameraProvider) {
cameraProvider.unbindAll();
CameraSelector cameraSelector = new CameraSelector.Builder()
.requireLensFacing(CameraSelector.LENS_FACING_BACK)
.build();
Preview preview = new Preview.Builder()
.build();
preview.setSurfaceProvider(previewView.getSurfaceProvider());
// Image capture use case
imageCapture = new ImageCapture.Builder()
.setCaptureMode(ImageCapture.CAPTURE_MODE_MINIMIZE_LATENCY)
.build();
cameraProvider.bindToLifecycle( this, cameraSelector, preview, imageCapture);
}
@SuppressLint("RestrictedApi")
@Override
public void onClick(View view) {
switch (view.getId()) {
case R.id.bCapture:
capturePhoto();
break;
}
}
private void capturePhoto() {
ContentValues contentValues = new ContentValues();
imageCapture.takePicture(
new ImageCapture.OutputFileOptions.Builder(
getContentResolver(),
MediaStore.Images.Media.EXTERNAL_CONTENT_URI,
contentValues
).build(),
getExecutor(),
new ImageCapture.OnImageSavedCallback() {
@Override
public void onImageSaved(@NonNull ImageCapture.OutputFileResults outputFileResults) {
Toast.makeText(MainActivity.this, "Photo has been saved successfully.", Toast.LENGTH_SHORT).show();
Bitmap image = BitmapFactory.decodeFile(getLastImageId());
int dimension = Math.min(image.getWidth(), image.getHeight());
image = ThumbnailUtils.extractThumbnail(image, dimension, dimension);
image = Bitmap.createScaledBitmap(image, 224, 224, false);
imageView.setImageBitmap(image);
classifyImage(image);
}
@Override
public void onError(@NonNull ImageCaptureException exception) {
Toast.makeText(MainActivity.this, "Error saving photo: " + exception.getMessage(), Toast.LENGTH_SHORT).show();
}
}
);
}
private String getLastImageId(){
final String[] imageColumns = { MediaStore.Images.Media._ID, MediaStore.Images.Media.DATA };
final String imageOrderBy = MediaStore.Images.Media._ID+" DESC";
Cursor imageCursor = managedQuery(MediaStore.Images.Media.EXTERNAL_CONTENT_URI, imageColumns, null, null, imageOrderBy);
if(imageCursor.moveToFirst()){
@SuppressLint("Range") String fullPath = imageCursor.getString(imageCursor.getColumnIndex(MediaStore.Images.Media.DATA));
imageCursor.close();
System.out.println(fullPath);
return fullPath;
}else{
return "";
}
}
private void classifyImage(Bitmap image) {
try {
Model model = Model.newInstance(getApplicationContext());
// Creates inputs for reference.
TensorBuffer inputFeature0 = TensorBuffer.createFixedSize(new int[]{1, 224, 224, 3}, DataType.FLOAT32);
ByteBuffer byteBuffer = ByteBuffer.allocateDirect(4*224*224*3);
byteBuffer.order(ByteOrder.nativeOrder());
int[] intValues = new int[224*224];
image.getPixels(intValues, 0, image.getWidth(), 0, 0, image.getWidth(), image.getHeight());
int pixel = 0;
for(int i = 0; i < 224; i++) {
int val = intValues[pixel++];
byteBuffer.putFloat(((val >> 16) & 0xFF) * (1.f/255.f));
byteBuffer.putFloat(((val >> 8) & 0xFF) * (1.f/255.f));
byteBuffer.putFloat(((val & 0xFF) * (1.f/255.f)));
}
inputFeature0.loadBuffer(byteBuffer);
// Runs model inference and gets result.
Model.Outputs outputs = model.process(inputFeature0);
TensorBuffer outputFeature0 = outputs.getOutputFeature0AsTensorBuffer();
float[] conifdences = outputFeature0.getFloatArray();
int maxPos = 0;
float maxConf = 0;
for(int i = 0; i < conifdences.length; i++) {
if(conifdences[i] > maxConf) {
maxConf = conifdences[i];
maxPos = i;
}
}
String[] classes = {"chicken", "apple", "things", "things"};
System.out.println(classes[maxPos]);
String s = "";
for(int i = 0; i < classes.length; i++) {
s+= String.format("%s: %.1f%%\n", classes[i], conifdences[i]*100);
}
System.out.println(s);
// Releases model resources if no longer used.
model.close();
}
}