feat: added logging feature

Author: Ahwar

README.md

@@ -244,6 +244,7 @@ optional arguments:
  suitable plugin for device specified (CPU by default)
 9. `-pt` Probability threshold for face detections filtering(0.5 by
  default)
+10. `-pr` Set to True if you want to preview visualizations on person face.
 ## Benchmarks
 
 ### Model Loading Time

src/face_detector.py

@@ -82,7 +82,7 @@ def predict(self, image):
  # Start Async Inference Request
  infer_request_handle = self.exec_network.start_async(
  request_id=0, inputs={self.input_blob: input_img})
- 
+
  # wait for the output and return.
  if infer_request_handle.wait(-1) == 0:
  model_output = infer_request_handle.outputs[self.output_blob]
@@ -130,7 +130,7 @@ def preprocess_output(self, outputs, threshold, image, image_w, image_h):
  (x_min, y_min) = (int(det[3] * image_w), int(det[4] * image_h))
  (x_max, y_max) = (int(det[5] * image_w), int(det[6] * image_h))
  cropped_face = image[y_min:y_max, x_min:x_max]
- return cropped_face
+ return cropped_face, ((x_min, y_min), (x_max, y_max))
 
  def get_input_shape(self):
  """ Return the shape of the input layer """

src/landmark_detector.py

@@ -121,13 +121,11 @@ def preprocess_output(self, outputs, image):
 
  x_left_eye, y_left_eye = outputs[0][0][0][0], outputs[0][1][0][0]
  x_right_eye, y_right_eye = outputs[0][2][0][0], outputs[0][3][0][0]
-
- left_eye = self.crop_eyes(x_left_eye, y_left_eye, image)
- right_eye = self.crop_eyes(x_right_eye, y_right_eye, image)
- # print(left_eye.shape, right_eye.shape)
- # cv2.imwrite("left.png", left_eye)
- # cv2.imwrite("right.png", right_eye)
- return left_eye, right_eye
+ # make cropped eye and its coordinates
+ left_eye, left_coords = self.crop_eyes(x_left_eye, y_left_eye, image)
+ right_eye, right_coords = self.crop_eyes(x_right_eye, y_right_eye,
+ image)
+ return left_eye, left_coords, right_eye, right_coords
 
  def crop_eyes(self, x_axis, y_axis, image):
  w, h = image.shape[1], image.shape[0]
@@ -137,7 +135,7 @@ def crop_eyes(self, x_axis, y_axis, image):
  y_max = int(y_axis * h) + 30
 
  cropped_eye = image[y_min:y_max, x_min:x_max]
- return cropped_eye
+ return cropped_eye, ((x_min, y_min), (x_max, y_max))
 
  def get_input_shape(self):
  """ Return the shape of the input layer """

src/main.py

@@ -87,6 +87,11 @@ def build_argparser():
  default=0.5,
  help="Probability threshold for face detections filtering"
  "(0.5 by default)")
+ parser.add_argument(
+ "-pr",
+ "--preview",
+ action='store_true',
+ help="Use this flag if you want to preview visualizations on person face")
  return parser
 
 
@@ -192,8 +197,8 @@ def infer_on_stream(args):
  ### Detect Face in Frame
  output = face_detector.predict(frame)
  ## Crop Face
- face = face_detector.preprocess_output(output, args.prob_threshold,
-   frame, initial_w, initial_h)
+ face, face_coords = face_detector.preprocess_output(
+ output, args.prob_threshold, frame, initial_w, initial_h)
 
  # skip frame if face not found
  if not np.any(face):
@@ -208,7 +213,7 @@ def infer_on_stream(args):
  landmarks = landmark_detector.predict(face)
  logger.info("Face Landmarks detected")
  ## Crop left and right Eye
- left_eye, right_eye = landmark_detector.preprocess_output(
+ left_eye, left_eye_coords, right_eye, right_eye_coords = landmark_detector.preprocess_output(
  landmarks, face)
 
  ## Skip frame if any eye is not cropped correctly
@@ -223,6 +228,46 @@ def infer_on_stream(args):
  ## Get mouse coords (x, y)
  mouse_coords = gaze_estimator.preprocess_output(gaze, head_pose)
  logger.info("New mouse coordinates: {}".format(mouse_coords))
+ # Show Preview of input with drawn predictions
+ if (args.preview):
+ # function draw rectangel around eye
+ def rectange_eyes(frame, face_coords, eye_coords):
+ """Draw bounding box around Eye"""
+ eye_start = (
+ (face_coords[0][0] + eye_coords[0][0]), # x_min + x_min
+ (face_coords[0][1] + eye_coords[0][1])) # y_min _ y_min
+ eye_end = (
+ (face_coords[0][0] + eye_coords[1][0]), # x_min + x_max
+ (face_coords[0][1] + eye_coords[1][1])) # y_min + y_max
+
+ return cv2.rectangle(frame, eye_start, eye_end, (0, 0, 255), 2)
+
+ # draw box around face
+ image = cv2.rectangle(frame, face_coords[0], face_coords[1],
+ (0, 0, 255), 2)
+ # draw box around left eye
+ image = rectange_eyes(image, face_coords, left_eye_coords)
+ # draw box around right eye
+ image = rectange_eyes(image, face_coords, right_eye_coords)
+ # show head pose values on image
+ cv2.putText(
+ image,
+ "Head Pose: Yaw: {:.2f}, Pitch: {:.2f}, Roll: {:.2f}".format(
+ head_pose["angle_y_fc"][0][0],
+ head_pose["angle_y_fc"][0][0],
+ head_pose["angle_y_fc"][0][0],
+ ), (40, 40), cv2.FONT_HERSHEY_COMPLEX, 0.8, (0, 255, 0), 1)
+ # show head pose values on image
+ cv2.putText(
+ image,
+ "Gaze: X-axis: {:.2f}, Y-axis: {:.2f}, Z-axis: {:.2f}".format(
+ gaze[0][0], gaze[0][1], gaze[0][2]), (40, 70),
+ cv2.FONT_HERSHEY_COMPLEX, 0.8, (0, 255, 0), 1)
+
+ cv2.imshow('Preview | press q to close', image)
+ # break loop if q is pressed on output window
+ if cv2.waitKey(1) & 0xFF == ord('q'):
+ break
 
  print("New mouse coordinates: {}\n\n".format(mouse_coords))
  ### Move Mouse