headcrop now based on width of the head

script/headdetector.py

@@ -9,7 +9,7 @@
  gazemachine = Gaze(model_def, model_weights)
  faceCascade=cv2.CascadeClassifier('/home/pieter/projects/engagement-l2tor/data/haarcascade_frontalface_alt.xml')
  print(faceCascade.empty())
- video = cv2.VideoCapture(0)
+ video = cv2.VideoCapture('/home/pieter/projects/engagement-l2tor/data/403059_les2_fragment1.mp4')
  while True:
  # Capture frame-by-frame
  #time.sleep(1)
@@ -21,17 +21,19 @@
  # Draw a rectangle around the faces
  x_face = 0
  y_face = 0
+ alpha = 0
  for (x, y, w, h) in faces:
  x_face = int((x + (x+w))/2)
  y_face = int((y+ (y+h))/2)
- cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
+ alpha = w / x_frame
+ #cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
  # Display the resulting frame
  e = [x_face/x_frame, y_face/y_frame,]
  image = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
  if e[0] != 0.0 and e[1] != 0.0:
- predictions = gazemachine.getGaze(image, e)
+ predictions = gazemachine.getGaze(image, e, alpha)
  cv2.imshow('Eyeframe', gazemachine.getEyeImage())
- #cv2.circle(frame, (predictions[0], predictions[1]), 10, (0, 255,0), 2)
+ cv2.circle(frame, (predictions[0], predictions[1]), 10, (0, 255,0), 2)
  cv2.line(frame, (x_face, y_face), (predictions[0], predictions[1]), (0, 255, 0), 2)
  print(predictions)
  cv2.imshow('Video', frame)

script/predict_gaze.py

@@ -23,7 +23,7 @@ def getHeatmap(self):
  """Get heatmap to see where the gaze is predicted."""
  return self.heatmap
 
- def getGaze(self, image, headloc):
+ def getGaze(self, image, headloc, alpha=0.3):
  """Returns x,y coordinates of the gaze location
 
  Keyword Arguments:
@@ -34,7 +34,7 @@ def getGaze(self, image, headloc):
  self.head_x = int(headloc[0] * np.shape(self.image)[1])
  self.head_y = int(headloc[1] * np.shape(self.image)[0])
  self.img_resize, self.eye_image_resize, self.z =\
- self.prepImages(self.image, self.headloc)
+ self.prepImages(self.image, self.headloc, alpha)
  self.network_outputs = self.predictGaze()
  self.heatmap, self.predictions =\
  self.postProcessing(self.network_outputs)
@@ -60,7 +60,7 @@ def visualizeGaze(self):
  ax.add_patch(Circle((self.head_x, self.head_y), 10, color = 'r'))
  plt.show()
 
- def prepImages(self, img, e):
+ def prepImages(self, img, e, alpha):
  """
  Output images of prepImages are exactly the same as the matlab ones
 
@@ -69,7 +69,6 @@ def prepImages(self, img, e):
  e -- head location (relative) [x, y]
  """
  input_shape = [227, 227]
- alpha = 0.3
  img_resize = None
 
  wy = int(alpha * img.shape[0])
@@ -80,6 +79,10 @@ def prepImages(self, img, e):
  x1 = int(center[0]-.5*wx) - 1
  x2 = int(center[0]+.5*wx) - 1
  #make crop of face from image
+ if y1 < 0:
+ y1 = 0
+ if x1 < 0:
+ x1 = 0
  im_face = img[y1:y2, x1:x2, :]
 
  #subtract mean from images