/**
* Process all the data in the training data set to learn the classifications. See code for
* details.
*/
public void learnAndSave() {
System.out.println("======== Learning Classifier");
// Either load pre-computed words or compute the words from the training images
AssignCluster<double[]> assignment;
if (new File(CLUSTER_FILE_NAME).exists()) {
assignment = UtilIO.load(CLUSTER_FILE_NAME);
} else {
System.out.println(" Computing clusters");
assignment = computeClusters();
}
// Use these clusters to assign features to words
FeatureToWordHistogram_F64 featuresToHistogram =
new FeatureToWordHistogram_F64(assignment, HISTOGRAM_HARD);
// Storage for the work histogram in each image in the training set and their label
List<HistogramScene> memory;
if (!new File(HISTOGRAM_FILE_NAME).exists()) {
System.out.println(" computing histograms");
memory = computeHistograms(featuresToHistogram);
UtilIO.save(memory, HISTOGRAM_FILE_NAME);
}
}
public static void main(String[] args) {
DescribeImageDense<ImageUInt8, TupleDesc_F64> desc =
(DescribeImageDense)
FactoryDescribeImageDense.surfFast(
null, new ConfigDenseSample(DESC_SCALE, DESC_SKIP, DESC_SKIP), ImageUInt8.class);
ComputeClusters<double[]> clusterer =
FactoryClustering.kMeans_F64(null, MAX_KNN_ITERATIONS, 20, 1e-6);
clusterer.setVerbose(true);
NearestNeighbor<HistogramScene> nn = FactoryNearestNeighbor.exhaustive();
ExampleClassifySceneKnn example = new ExampleClassifySceneKnn(desc, clusterer, nn);
File trainingDir = new File(UtilIO.pathExample("learning/scene/train"));
File testingDir = new File(UtilIO.pathExample("learning/scene/test"));
if (!trainingDir.exists() || !testingDir.exists()) {
System.err.println(
"Please follow instructions in data/applet/learning/scene and download the");
System.err.println("required files");
System.exit(1);
}
example.loadSets(trainingDir, null, testingDir);
// train the classifier
example.learnAndSave();
// now load it for evaluation purposes from the files
example.loadAndCreateClassifier();
// test the classifier on the test set
Confusion confusion = example.evaluateTest();
confusion.getMatrix().print();
System.out.println("Accuracy = " + confusion.computeAccuracy());
// Show confusion matrix
// Not the best coloration scheme... perfect = red diagonal and blue elsewhere.
ShowImages.showWindow(
new ConfusionMatrixPanel(confusion.getMatrix(), 400, true), "Confusion Matrix", true);
// For "fast" SURF descriptor the accuracy is 52.2%
// For "stable" SURF descriptor the accuracy is 49.4%
// This is interesting. When matching images "stable" is significantly better than "fast"
// One explanation is that the descriptor for "fast" samples a smaller region than "stable", by
// a
// couple of pixels at scale of 1. Thus there is less overlap between the features.
// Reducing the size of "stable" to 0.95 does slightly improve performance to 50.5%, can't scale
// it down
// much more without performance going down
}
/**
* Extract dense features across the training set. Then clusters are found within those features.
*/
private AssignCluster<double[]> computeClusters() {
System.out.println("Image Features");
// computes features in the training image set
features.reset();
for (String scene : train.keySet()) {
List<String> imagePaths = train.get(scene);
System.out.println(" " + scene);
for (String path : imagePaths) {
ImageUInt8 image = UtilImageIO.loadImage(path, ImageUInt8.class);
describeImage.process(image, features, null);
}
}
// add the features to the overall list which the clusters will be found inside of
for (int i = 0; i < features.size; i++) {
cluster.addReference(features.get(i));
}
System.out.println("Clustering");
// Find the clusters. This can take a bit
cluster.process(NUMBER_OF_WORDS);
UtilIO.save(cluster.getAssignment(), CLUSTER_FILE_NAME);
return cluster.getAssignment();
}
public void loadAndCreateClassifier() {
// load results from a file
List<HistogramScene> memory = UtilIO.load(HISTOGRAM_FILE_NAME);
AssignCluster<double[]> assignment = UtilIO.load(CLUSTER_FILE_NAME);
FeatureToWordHistogram_F64 featuresToHistogram =
new FeatureToWordHistogram_F64(assignment, HISTOGRAM_HARD);
// Provide the training results to K-NN and it will preprocess these results for quick lookup
// later on
// Can use this classifier with saved results and avoid the
classifier =
new ClassifierKNearestNeighborsBow<ImageUInt8, TupleDesc_F64>(
nn, describeImage, featuresToHistogram);
classifier.setClassificationData(memory, getScenes().size());
classifier.setNumNeighbors(NUM_NEIGHBORS);
}
public static void main(String args[]) {
DemoBinaryImageOpsApp app = new DemoBinaryImageOpsApp(GrayF32.class);
java.util.List<PathLabel> inputs = new ArrayList<>();
inputs.add(new PathLabel("particles", UtilIO.pathExample("particles01.jpg")));
inputs.add(new PathLabel("shapes", UtilIO.pathExample("shapes/shapes01.png")));
app.setInputList(inputs);
// wait for it to process one image so that the size isn't all screwed up
while (!app.getHasProcessedImage()) {
Thread.yield();
}
ShowImages.showWindow(app, "Binary Image Ops", true);
System.out.println("Done");
}
public static void main(String[] args) {
// Example with a moving camera. Highlights why motion estimation is sometimes required
String fileName = UtilIO.pathExample("tracking/chipmunk.mjpeg");
// Camera has a bit of jitter in it. Static kinda works but motion reduces false positives
// String fileName = UtilIO.pathExample("background/horse_jitter.mp4");
// Comment/Uncomment to switch input image type
ImageType imageType = ImageType.single(GrayF32.class);
// ImageType imageType = ImageType.il(3, InterleavedF32.class);
// ImageType imageType = ImageType.il(3, InterleavedU8.class);
// Configure the feature detector
ConfigGeneralDetector confDetector = new ConfigGeneralDetector();
confDetector.threshold = 10;
confDetector.maxFeatures = 300;
confDetector.radius = 6;
// Use a KLT tracker
PointTracker tracker =
FactoryPointTracker.klt(new int[] {1, 2, 4, 8}, confDetector, 3, GrayF32.class, null);
// This estimates the 2D image motion
ImageMotion2D<GrayF32, Homography2D_F64> motion2D =
FactoryMotion2D.createMotion2D(
500, 0.5, 3, 100, 0.6, 0.5, false, tracker, new Homography2D_F64());
ConfigBackgroundBasic configBasic = new ConfigBackgroundBasic(30, 0.005f);
// Configuration for Gaussian model. Note that the threshold changes depending on the number of
// image bands
// 12 = gray scale and 40 = color
ConfigBackgroundGaussian configGaussian = new ConfigBackgroundGaussian(12, 0.001f);
configGaussian.initialVariance = 64;
configGaussian.minimumDifference = 5;
// Comment/Uncomment to switch background mode
BackgroundModelMoving background =
FactoryBackgroundModel.movingBasic(
configBasic, new PointTransformHomography_F32(), imageType);
// FactoryBackgroundModel.movingGaussian(configGaussian, new PointTransformHomography_F32(),
// imageType);
MediaManager media = DefaultMediaManager.INSTANCE;
SimpleImageSequence video = media.openVideo(fileName, background.getImageType());
// media.openCamera(null,640,480,background.getImageType());
// ====== Initialize Images
// storage for segmented image. Background = 0, Foreground = 1
GrayU8 segmented = new GrayU8(video.getNextWidth(), video.getNextHeight());
// Grey scale image that's the input for motion estimation
GrayF32 grey = new GrayF32(segmented.width, segmented.height);
// coordinate frames
Homography2D_F32 firstToCurrent32 = new Homography2D_F32();
Homography2D_F32 homeToWorld = new Homography2D_F32();
homeToWorld.a13 = grey.width / 2;
homeToWorld.a23 = grey.height / 2;
// Create a background image twice the size of the input image. Tell it that the home is in the
// center
background.initialize(grey.width * 2, grey.height * 2, homeToWorld);
BufferedImage visualized =
new BufferedImage(segmented.width, segmented.height, BufferedImage.TYPE_INT_RGB);
ImageGridPanel gui = new ImageGridPanel(1, 2);
gui.setImages(visualized, visualized);
ShowImages.showWindow(gui, "Detections", true);
double fps = 0;
double alpha = 0.01; // smoothing factor for FPS
while (video.hasNext()) {
ImageBase input = video.next();
long before = System.nanoTime();
GConvertImage.convert(input, grey);
if (!motion2D.process(grey)) {
throw new RuntimeException("Should handle this scenario");
}
Homography2D_F64 firstToCurrent64 = motion2D.getFirstToCurrent();
UtilHomography.convert(firstToCurrent64, firstToCurrent32);
background.segment(firstToCurrent32, input, segmented);
background.updateBackground(firstToCurrent32, input);
long after = System.nanoTime();
fps = (1.0 - alpha) * fps + alpha * (1.0 / ((after - before) / 1e9));
VisualizeBinaryData.renderBinary(segmented, false, visualized);
gui.setImage(0, 0, (BufferedImage) video.getGuiImage());
gui.setImage(0, 1, visualized);
gui.repaint();
System.out.println("FPS = " + fps);
try {
Thread.sleep(5);
} catch (InterruptedException e) {
}
}
}