fixed all the examples and added an MNIST detector example

build.gradle

@@ -1,5 +1,5 @@
 group 'ch.bildspur'
-version '0.3.1'
+version '0.3.2'
 
 apply plugin: 'java'
 

examples/FaceDetectorSelfie/FaceDetectorSelfie.pde

@@ -0,0 +1,51 @@
+import ch.bildspur.vision.*;
+import ch.bildspur.vision.result.*;
+import java.util.List;
+
+PImage testImage;
+
+DeepVision vision;
+ULFGFaceDetectionNetwork network;
+List<ObjectDetectionResult> detections;
+
+public void setup() {
+  size(640, 480);
+  colorMode(HSB, 360, 100, 100);
+
+  vision = new DeepVision(this);
+  testImage = loadImage(sketchPath("data/selfie.jpg"));
+
+  println("creating network...");
+  network = vision.createULFGFaceDetectorRFB640();
+  network.setConfidenceThreshold(0.2);
+
+  println("loading model...");
+  network.setup();
+
+  //network.setConfidenceThreshold(0.2f);
+
+  println("inferencing...");
+  detections = network.run(testImage);
+  println("done!");
+
+  for (ObjectDetectionResult detection : detections) {
+    System.out.println(detection.getClassName() + "\t[" + detection.getConfidence() + "]");
+  }
+
+  println("found " + detections.size() + " faces!");
+}
+
+public void draw() {
+  background(55);
+  image(testImage, 0, 0);
+
+  noFill();
+  strokeWeight(2f);
+
+  stroke(200, 80, 100);
+  for (ObjectDetectionResult detection : detections) {
+    rect(detection.getX(), detection.getY(), detection.getWidth(), detection.getHeight());
+  }
+
+  surface.setTitle("Face Recognition Test - Faces: " + detections.size());
+}

examples/HumanPoseWebcam/HumanPoseWebcam.pde

@@ -11,6 +11,8 @@ HumanPoseResult result;
 
 Capture cam;
 
+float threshold = 0.5;
+
 void setup() {
   size(640, 480, FX2D);
 
@@ -81,6 +83,9 @@ private void drawHuman(HumanPoseResult human) {
   int i = 0;
   fill(0);
   for (KeyPointResult point : human.getKeyPoints()) {
+    if(point.getProbability() < threshold)
+      continue;
+
     ellipse(point.getX(), point.getY(), 10, 10);
     text(i, point.getX() + 5, point.getY());
     i++;
@@ -92,6 +97,9 @@ private void connect(KeyPointResult... keyPoints) {
     KeyPointResult a = keyPoints[i];
     KeyPointResult b = keyPoints[i + 1];
 
+    if(a.getProbability() < threshold || b.getProbability() < threshold)
+      continue;
+
     line(a.getX(), a.getY(), b.getX(), b.getY());
   }
 }

examples/MNISTDetector/MNISTDetector.pde

@@ -0,0 +1,93 @@
+import ch.bildspur.vision.*;
+import ch.bildspur.vision.result.*;
+
+DeepVision vision;
+MNISTClassificationNetwork network;
+
+ClassificationResult result;
+long inferenceTime = 0;
+PImage canvas = new PImage(28, 28, RGB);
+
+boolean mouseDrawing = false;
+float factor = 560 / 28;
+
+public void setup() {
+  size(560, 560, FX2D);
+  colorMode(HSB, 360, 100, 100);
+
+  vision = new DeepVision(this);
+
+  println("creating network...");
+  network = vision.createMNISTClassifier();
+
+  println("loading model...");
+  network.setup();
+
+  clearCanvas();
+  println("ready!");
+}
+
+public void draw() {
+  background(55);
+
+  if (mouseDrawing) {
+    int x = round(mouseX / factor);
+    int y = round(mouseY / factor);
+
+    canvas.set(x, y, color(255));
+    canvas.updatePixels();
+  }
+
+  //image(canvas, 0, 0, width, height);
+  for (int y = 0; y < canvas.width; y++) {
+    for (int x = 0; x < canvas.width; x++) {
+      color c = canvas.get(x, y);
+      fill(c);
+      noStroke();
+      rect(x * factor, y * factor, factor, factor);
+    }
+  }
+
+  // display info
+  fill(140, 80, 100);
+  textSize(16);
+  text("space: run inference / c: clear canvas", 10, 20);
+
+  if (result != null) {
+    text("Detected '" + result.getClassName() + "' with " + result.getConfidence() + "%", 10, 50);
+  }
+
+  surface.setTitle("MNIST Detector");
+}
+
+void mousePressed() {
+  mouseDrawing = true;
+}
+
+void mouseReleased() {
+  mouseDrawing = false;
+}
+
+void keyPressed() {
+  if (key == ' ') {
+    println("inferencing...");
+    int start = millis();
+    result = network.run(canvas);
+    inferenceTime = millis() - start;
+  }
+
+  if (key == 'c' || key == 'C') {
+    println("clearing canvas...");
+    clearCanvas();
+    result = null;
+  }
+}
+
+void clearCanvas() {
+  for (int y = 0; y < canvas.width; y++) {
+    for (int x = 0; x < canvas.width; x++) {
+      canvas.set(x, y, color(0));
+    }
+  }
+  canvas.updatePixels();
+}

examples/RealSensePoseCapture/RealSensePoseCapture.pde

@@ -1,5 +1,5 @@
 import ch.bildspur.vision.DeepVision;
-import ch.bildspur.vision.SingleHumanPoseEstimationNetwork;
+import ch.bildspur.vision.SingleHumanPoseNetwork;
 import ch.bildspur.vision.result.HumanPoseResult;
 import ch.bildspur.vision.result.KeyPointResult;
 
@@ -9,7 +9,7 @@ RealSenseCamera camera = new RealSenseCamera(this);
 
 DeepVision vision = new DeepVision(this);
 
-SingleHumanPoseEstimationNetwork pose;
+SingleHumanPoseNetwork pose;
 HumanPoseResult result;
 
 void setup()

examples/RealSenseYoloDetector/RealSenseYoloDetector.pde

@@ -53,7 +53,7 @@ void draw()
   strokeWeight(2f);
 
   for (ObjectDetectionResult detection : result) {
-    stroke(round(360.0f * (float) detection.getClassId() / net.getNames().size()), 75, 100);
+    stroke(round(360.0f * (float) detection.getClassId() / net.getClassNames().size()), 75, 100);
     rect(detection.getX(), detection.getY(), detection.getWidth(), detection.getHeight());
 
     textSize(15);

examples/SingleHumanPoseEstimation/SingleHumanPoseEstimation.pde

@@ -5,9 +5,9 @@ import ch.bildspur.vision.result.KeyPointResult;
 
 PImage testImage;
 
-DeepVision vision = new DeepVision(this);
+DeepVision vision;
 
-SingleHumanPoseEstimationNetwork pose;
+SingleHumanPoseNetwork pose;
 HumanPoseResult result;
 
 PImage image;
@@ -16,10 +16,10 @@ void setup() {
   size(480, 640, FX2D);
 
   colorMode(HSB, 360, 100, 100);
-
   testImage = loadImage(sketchPath("data/pose.jpg"));
 
   println("creating network...");
+  vision = new DeepVision(this);
   pose = vision.createSingleHumanPoseEstimation();
 
   println("loading model...");

examples/YOLODetectObjects/YOLODetectObjects.pde

@@ -16,25 +16,26 @@ public void setup() {
   image = loadImage("hk.jpg");
 
   println("creating model...");
-  yolo = deepVision.createYOLOv3Tiny();
+  yolo = deepVision.createYOLOv3();
 
   println("loading yolo model...");
   yolo.setup();
+
+  println("inferencing...");
+  yolo.setConfidenceThreshold(0.3f);
+  detections = yolo.run(image);
 }
 
 public void draw() {
   background(55);
 
   image(image, 0, 0);
 
-  yolo.setConfidenceThreshold(0.2f);
-  detections = yolo.run(image);
-
   noFill();
   strokeWeight(2f);
 
   for (ObjectDetectionResult detection : detections) {
-    stroke((int)(360.0 / yolo.getNames().size() * detection.getClassId()), 80, 100);
+    stroke((int)(360.0 / yolo.getClassNames().size() * detection.getClassId()), 80, 100);
     rect(detection.getX(), detection.getY(), detection.getWidth(), detection.getHeight());
   }
 

examples/YoloWecamExample/YoloWecamExample.pde

@@ -25,7 +25,8 @@ public void setup() {
   println("loading yolo model...");
   yolo.setup();
 
-  cam = new Capture(this);
+  String[] cams = Capture.list();
+  cam = new Capture(this, cams[0]);
   cam.start();
 }
 
@@ -47,7 +48,7 @@ public void draw() {
   strokeWeight(2f);
 
   for (ObjectDetectionResult detection : detections) {
-    stroke((int)(360.0 / yolo.getNames().size() * detection.getClassId()), 80, 100);
+    stroke((int)(360.0 / yolo.getClassNames().size() * detection.getClassId()), 80, 100);
     rect(detection.getX(), detection.getY(), detection.getWidth(), detection.getHeight());
   }
 

library.properties

@@ -4,7 +4,7 @@ authors=[Florian Bruggisser](https://broox.ch)
 url=https://github.com/cansik/deep-vision-processing
 sentence=Deep computer-vision algorithms for Processing
 paragraph=Run deep neural networks on images in Processing.
-version=003010
-prettyVersion=0.3.1
+version=003020
+prettyVersion=0.3.2
 minRevision=0
 maxRevision=0

release.sh

@@ -45,6 +45,11 @@ cp "build/libs/$ARCHIVE_NAME-complete.jar" "$OUTPUT/library/$ARCHIVE_NAME.jar"
 # cp -r native "$OUTPUT/library/"
 cp -r "build/docs/javadoc" "$OUTPUT/reference"
 
+# clean networks from examples
+cd "examples"
+rm -rf **/networks
+cd ..
+
 cp -r "examples" "$OUTPUT/"
 cp library.properties "$OUTPUT/"
 cp -r readme "$OUTPUT/"

src/main/java/ch/bildspur/vision/MNISTClassificationNetwork.java

@@ -67,7 +67,8 @@ public ClassificationResult run(Mat frame) {
         for (int i = 0; i < out.cols(); i++) {
             float probability = data.get(i);
 
-            System.out.println("# " + i + ": " + probability + "%");
+            // todo: fix probability issue
+            // System.out.println("# " + i + ": " + probability + "%");
 
             if (probability > maxProbability) {
                 maxProbability = probability;

src/test/java/ch/bildspur/vision/test/FaceRecognitionTest.java

@@ -32,7 +32,7 @@ public void setup() {
         testImage = loadImage(sketchPath("data/office.jpg"));
 
         println("creating network...");
-        network = vision.createULFGFaceDetectorSlim640();
+        network = vision.createULFGFaceDetectorRFB640();
 
         println("loading model...");
         network.setup();