This paper documents for pb, detection and identification goals, NDK, and does not require mixing CMake compiled programming c / c ++ file Android TensorFlow API
Only Module is build.gradle input module in Android project
// Tensorflow
compile 'org.tensorflow:tensorflow-android:1.13.1'Since many can be found at both the source code, it is not an additional project, just the recording pit mining of the learning process
Usually images need tested, locating and identifying
A mode : Single Target Recognition
This way you can take pictures by phone or select a picture album, manually select the target crop, the target identified in the model into
Use of pre-training model is mainly inception frame v1, v2, v3, etc.
Own data model training tools can be downloaded:
https://github.com/tensorflow/hub/blob/master/examples/image_retraining/retrain.pyTrain yourself to be a reference model of the process:
https://www.tensorflow.org/hub/tutorials/image_retrainingAndroid end of the main call interface
/**
* A classifier specialized to label images using TensorFlow.
*/
public class TensorFlowImageClassifier implements Classifier {
private static final String TAG = "ImageClassifier";
// Only return this many results with at least this confidence.
private static final int MAX_RESULTS = 3;
private static final float THRESHOLD = 0.3f;
// Config values.
private String inputName;
private String outputName;
private int inputSize;
private int imageMean;
private float imageStd;
// Pre-allocated buffers.
private Vector<String> labels = new Vector<String>();
private int[] intValues;
private float[] floatValues;
private float[] outputs;
private String[] outputNames;
private TensorFlowInferenceInterface inferenceInterface;
private boolean runStats = false;
private TensorFlowImageClassifier() {
}
/**
* Initializes a native TensorFlow session for classifying images.
*
* @param assetManager The asset manager to be used to load assets.
* @param modelFilename The filepath of the model GraphDef protocol buffer.
* @param labelFilename The filepath of label file for classes.
* @param inputSize The input size. A square image of inputSize x inputSize is assumed.
* @param imageMean The assumed mean of the image values.
* @param imageStd The assumed std of the image values.
* @param inputName The label of the image input node.
* @param outputName The label of the output node.
* @throws IOException
*/
public static Classifier create(
AssetManager assetManager,
String modelFilename,
String labelFilename,
int inputSize,
int imageMean,
float imageStd,
String inputName,
String outputName)
throws IOException {
TensorFlowImageClassifier c = new TensorFlowImageClassifier();
c.inputName = inputName;
c.outputName = outputName;
// Read the label names into memory.
// TODO(andrewharp): make this handle non-assets.
//read labels for label file
c.labels = readLabels(assetManager, labelFilename);
c.inferenceInterface = new TensorFlowInferenceInterface(assetManager, modelFilename);
// The shape of the output is [N, NUM_CLASSES], where N is the batch size.
int numClasses = (int) c.inferenceInterface.graph().operation(outputName).output(0).shape().size(1);
Log.i(TAG, "Read " + c.labels.size() + " labels, output layer size is " + numClasses);
// Ideally, inputSize could have been retrieved from the shape of the input operation. Alas,
// the placeholder node for input in the graphdef typically used does not specify a shape, so it
// must be passed in as a parameter.
c.inputSize = inputSize;
c.imageMean = imageMean;
c.imageStd = imageStd;
// Pre-allocate buffers.
c.outputNames = new String[]{outputName};
c.intValues = new int[inputSize * inputSize];
c.floatValues = new float[inputSize * inputSize * 3];
c.outputs = new float[numClasses];
return c;
}
//given a saved drawn model, lets read all the classification labels that are
//stored and write them to our in memory labels list
private static Vector<String> readLabels(AssetManager am, String fileName) throws IOException {
BufferedReader br = new BufferedReader(new InputStreamReader(am.open(fileName)));
String line;
Vector<String> labels = new Vector<>();
while ((line = br.readLine()) != null) {
labels.add(line);
}
br.close();
return labels;
}
@Override
public List<Recognition> recognizeImage(final Bitmap bitmap) {
// Log this method so that it can be analyzed with systrace.
TraceCompat.beginSection("recognizeImage");
TraceCompat.beginSection("preprocessBitmap");
// Preprocess the image data from 0-255 int to normalized float based
// on the provided parameters.
bitmap.getPixels(intValues, 0, bitmap.getWidth(), 0, 0, bitmap.getWidth(), bitmap.getHeight());
for (int i = 0; i < intValues.length; ++i) {
final int val = intValues[i];
floatValues[i * 3 + 0] = (((val >> 16) & 0xFF) - imageMean) / imageStd;
floatValues[i * 3 + 1] = (((val >> 8) & 0xFF) - imageMean) / imageStd;
floatValues[i * 3 + 2] = ((val & 0xFF) - imageMean) / imageStd;
}
TraceCompat.endSection();
// Copy the input data into TensorFlow.
TraceCompat.beginSection("feed");
inferenceInterface.feed(inputName, floatValues, new long[]{1, inputSize, inputSize, 3});
// inferenceInterface.feed("is_training",new boolean[]{false});
// inferenceInterface.feed("keep_prob",new float[]{1.0f});
TraceCompat.endSection();
// Run the inference call.
TraceCompat.beginSection("run");
inferenceInterface.run(outputNames, runStats);
TraceCompat.endSection();
// Copy the output Tensor back into the output array.
TraceCompat.beginSection("fetch");
inferenceInterface.fetch(outputName, outputs);
TraceCompat.endSection();
// Find the best classifications.
PriorityQueue<Recognition> pq =
new PriorityQueue<Recognition>(
3,
new Comparator<Recognition>() {
@Override
public int compare(Recognition lhs, Recognition rhs) {
// Intentionally reversed to put high confidence at the head of the queue.
return Float.compare(rhs.getConfidence(), lhs.getConfidence());
}
});
for (int i = 0; i < outputs.length; ++i) {
if (outputs[i] > THRESHOLD) {
pq.add(
new Recognition("" + i, labels.size() > i ? labels.get(i) : "unknown", outputs[i], null));
}
}
final ArrayList<Recognition> recognitions = new ArrayList<Recognition>();
int recognitionsSize = Math.min(pq.size(), MAX_RESULTS);
for (int i = 0; i < recognitionsSize; ++i) {
recognitions.add(pq.poll());
}
TraceCompat.endSection(); // "recognizeImage"
return recognitions;
}Pit Point:
- Different frame models, different training model input placeholder must correspond;
TraceCompat.beginSection("feed");
inferenceInterface.feed(inputName, floatValues, new long[]{1, inputSize, inputSize, 3});
// inferenceInterface.feed("is_training",new boolean[]{false});
// inferenceInterface.feed("keep_prob",new float[]{1.0f});
TraceCompat.endSection();
Can be viewed using Android studio pb file, usually the first few lines will tell you whether is_training, keep_prob and other placeholder, if any, must join in the Android program.
- Training data set must align, resize and consistent framework image size model, Android end call interface, input parameters and the training images must be consistent, otherwise the classification error. Personal speculation: different types of image footprint size is the same, should the image varies, can lead to confusion pb map vector data.
private static final int FACE_SIZE = 299;
private static final int IMAGE_MEAN = 128;
private static final float IMAGE_STD = 128;
- The generated model can not be placed directly into Android, the need to step conversion: the official explanation:
To use v3 Inception model, strip the DecodeJpeg Op from your retrained
// model first:cd to D: \ Program Files \ Anaconda3 \ Lib \ site-packages \ tensorflow \ python \ tools directory, copy the step of generating a modified file to the directory output_graph.pb, execute the command :
python strip_unused.py --input_graph=face.pb --input_binary=true --output_graph=facenet.pb --input_node_names="Mul" --output_node_names="output"
- Android platform SDK is greater than 24 or more, call the camera or crop the picture will appear flash back phenomenon, to fear, usually caused by unsafe path Android7.0 is prohibited to access external, data is temporarily stored lead, you can network to find solutions to this problem;
Method two: target detection and identification
This embodiment can be camera, camera, or select an album, without cropping the image, directly into the model for the target object detection, location and identification
Use of pre-training model is mainly ssd_inception_v1, v2, v3 and other frameworks
Own data model training tools can be downloaded:
https://github.com/tensorflow/modelsTrain yourself to be a reference model of the process:
https://blog.csdn.net/zj1131190425/article/details/80711857
public class TensorFlowObjectDetectionAPIModel implements Classifier {
private static final Logger LOGGER = new Logger();
// Only return this many results.
private static final int MAX_RESULTS = 100;
// Config values.
private String inputName;
private int inputSize;
// Pre-allocated buffers.
private Vector<String> labels = new Vector<String>();
private int[] intValues;
private byte[] byteValues;
private float[] outputLocations;
private float[] outputScores;
private float[] outputClasses;
private float[] outputNumDetections;
private String[] outputNames;
private boolean logStats = false;
private TensorFlowInferenceInterface inferenceInterface;
/**
* Initializes a native TensorFlow session for classifying images.
*
* @param assetManager The asset manager to be used to load assets.
* @param modelFilename The filepath of the model GraphDef protocol buffer.
* @param labelFilename The filepath of label file for classes.
*/
public static Classifier create(
final AssetManager assetManager,
final String modelFilename,
final String labelFilename,
final int inputSize) throws IOException {
final TensorFlowObjectDetectionAPIModel d = new TensorFlowObjectDetectionAPIModel();
InputStream labelsInput = null;
String actualFilename = labelFilename.split("file:///android_asset/")[1];
labelsInput = assetManager.open(actualFilename);
BufferedReader br = null;
br = new BufferedReader(new InputStreamReader(labelsInput));
String line;
while ((line = br.readLine()) != null) {
LOGGER.w(line);
d.labels.add(line);
}
br.close();
d.inferenceInterface = new TensorFlowInferenceInterface(assetManager, modelFilename);
final Graph g = d.inferenceInterface.graph();
d.inputName = "image_tensor";
// The inputName node has a shape of [N, H, W, C], where
// N is the batch size
// H = W are the height and width
// C is the number of channels (3 for our purposes - RGB)
final Operation inputOp = g.operation(d.inputName);
if (inputOp == null) {
throw new RuntimeException("Failed to find input Node '" + d.inputName + "'");
}
d.inputSize = inputSize;
// The outputScoresName node has a shape of [N, NumLocations], where N
// is the batch size.
final Operation outputOp1 = g.operation("detection_scores");
if (outputOp1 == null) {
throw new RuntimeException("Failed to find output Node 'detection_scores'");
}
final Operation outputOp2 = g.operation("detection_boxes");
if (outputOp2 == null) {
throw new RuntimeException("Failed to find output Node 'detection_boxes'");
}
final Operation outputOp3 = g.operation("detection_classes");
if (outputOp3 == null) {
throw new RuntimeException("Failed to find output Node 'detection_classes'");
}
// Pre-allocate buffers.
d.outputNames = new String[] {"detection_boxes", "detection_scores",
"detection_classes", "num_detections"};
d.intValues = new int[d.inputSize * d.inputSize];
d.byteValues = new byte[d.inputSize * d.inputSize * 3];
d.outputScores = new float[MAX_RESULTS];
d.outputLocations = new float[MAX_RESULTS * 4];
d.outputClasses = new float[MAX_RESULTS];
d.outputNumDetections = new float[1];
return d;
}
private TensorFlowObjectDetectionAPIModel() {}
@Override
public List<Recognition> recognizeImage(final Bitmap bitmap) {
// Log this method so that it can be analyzed with systrace.
Trace.beginSection("recognizeImage");
Trace.beginSection("preprocessBitmap");
// Preprocess the image data from 0-255 int to normalized float based
// on the provided parameters.
bitmap.getPixels(intValues, 0, bitmap.getWidth(), 0, 0, bitmap.getWidth(), bitmap.getHeight());
for (int i = 0; i < intValues.length; ++i) {
byteValues[i * 3 + 2] = (byte) (intValues[i] & 0xFF);
byteValues[i * 3 + 1] = (byte) ((intValues[i] >> 8) & 0xFF);
byteValues[i * 3 + 0] = (byte) ((intValues[i] >> 16) & 0xFF);
}
Trace.endSection(); // preprocessBitmap
// Copy the input data into TensorFlow.
Trace.beginSection("feed");
inferenceInterface.feed(inputName, byteValues, 1, inputSize, inputSize, 3);
Trace.endSection();
// Run the inference call.
Trace.beginSection("run");
inferenceInterface.run(outputNames, logStats);
Trace.endSection();
// Copy the output Tensor back into the output array.
Trace.beginSection("fetch");
outputLocations = new float[MAX_RESULTS * 4];
outputScores = new float[MAX_RESULTS];
outputClasses = new float[MAX_RESULTS];
outputNumDetections = new float[1];
inferenceInterface.fetch(outputNames[0], outputLocations);
inferenceInterface.fetch(outputNames[1], outputScores);
inferenceInterface.fetch(outputNames[2], outputClasses);
inferenceInterface.fetch(outputNames[3], outputNumDetections);
Trace.endSection();
// Find the best detections.
final PriorityQueue<Recognition> pq =
new PriorityQueue<Recognition>(
1,
new Comparator<Recognition>() {
@Override
public int compare(final Recognition lhs, final Recognition rhs) {
// Intentionally reversed to put high confidence at the head of the queue.
return Float.compare(rhs.getConfidence(), lhs.getConfidence());
}
});
// Scale them back to the input size.
for (int i = 0; i < outputScores.length; ++i) {
final RectF detection =
new RectF(
outputLocations[4 * i + 1] * inputSize,
outputLocations[4 * i] * inputSize,
outputLocations[4 * i + 3] * inputSize,
outputLocations[4 * i + 2] * inputSize);
pq.add(
new Recognition("" + i, labels.get((int) outputClasses[i]), outputScores[i], detection));
}
final ArrayList<Recognition> recognitions = new ArrayList<Recognition>();
for (int i = 0; i < Math.min(pq.size(), MAX_RESULTS); ++i) {
recognitions.add(pq.poll());
}
Trace.endSection(); // "recognizeImage"
return recognitions;
}
@Override
public void enableStatLogging(final boolean logStats) {
this.logStats = logStats;
}
@Override
public String getStatString() {
return inferenceInterface.getStatString();
}
@Override
public void close() {
inferenceInterface.close();
}
}Pit Point:
- Copy the label .txt files in the assets directory
The first line is ???
Do not move, do not delete
-
labelImg data set must not duplicate names
- id and tag name label.pbtxt file, within tfrecord necessarily correspond to document classification id and name tags
