Reduced the impact of eye blinks and opening the mouth on head pose a…

…nd added new models to benchmark.

facetracker.py

@@ -134,8 +134,8 @@ def flush(self):
     model_base_path = get_model_base_path(args.model_dir)
     im = cv2.imread(os.path.join(model_base_path, "benchmark.bin"), cv2.IMREAD_COLOR)
     results = []
-    for model_type in [3, 2, 1, 0, -1, -2]:
-        tracker = Tracker(224, 224, threshold=0.1, max_threads=args.max_threads, max_faces=1, discard_after=0, scan_every=0, silent=True, model_type=model_type, model_dir=args.model_dir, no_gaze=(model_type < 0), detection_threshold=0.1, use_retinaface=0, max_feature_updates=900, static_model=True if args.no_3d_adapt == 1 else False)
+    for model_type in [3, 2, 1, 0, -1, -2, -3]:
+        tracker = Tracker(224, 224, threshold=0.1, max_threads=args.max_threads, max_faces=1, discard_after=0, scan_every=0, silent=True, model_type=model_type, model_dir=args.model_dir, no_gaze=(model_type == -1), detection_threshold=0.1, use_retinaface=0, max_feature_updates=900, static_model=True if args.no_3d_adapt == 1 else False)
         tracker.detected = 1
         tracker.faces = [(0, 0, 224, 224)]
         total = 0.0
@@ -245,7 +245,7 @@ def flush(self):
             first = False
             height, width, channels = frame.shape
             sock = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
-            tracker = Tracker(width, height, threshold=args.threshold, max_threads=args.max_threads, max_faces=args.faces, discard_after=args.discard_after, scan_every=args.scan_every, silent=False if args.silent == 0 else True, model_type=args.model, model_dir=args.model_dir, no_gaze=False if args.gaze_tracking != 0 and args.model >= 0 else True, detection_threshold=args.detection_threshold, use_retinaface=args.scan_retinaface, max_feature_updates=args.max_feature_updates, static_model=True if args.no_3d_adapt == 1 else False, try_hard=args.try_hard == 1)
+            tracker = Tracker(width, height, threshold=args.threshold, max_threads=args.max_threads, max_faces=args.faces, discard_after=args.discard_after, scan_every=args.scan_every, silent=False if args.silent == 0 else True, model_type=args.model, model_dir=args.model_dir, no_gaze=False if args.gaze_tracking != 0 and args.model != -1 else True, detection_threshold=args.detection_threshold, use_retinaface=args.scan_retinaface, max_feature_updates=args.max_feature_updates, static_model=True if args.no_3d_adapt == 1 else False, try_hard=args.try_hard == 1)
             if not args.video_out is None:
                 out = cv2.VideoWriter(args.video_out, cv2.VideoWriter_fourcc('F','F','V','1'), args.video_fps, (width * args.video_scale, height * args.video_scale))
 

tracker.py

@@ -328,10 +328,10 @@ def __init__(self, id, tracker):
         self.id = id
         self.frame_count = -1
         self.tracker = tracker
-        self.contour_pts = [0,1,8,15,16,27,28,29,30,31,32,33,34,35,36,39,42,45]
+        self.contour_pts = [0,1,15,16,27,28,29,30,31,32,33,34,35]
         self.face_3d = copy.copy(self.tracker.face_3d)
         if self.tracker.model_type == -1:
-            self.contour_pts = [0,2,8,14,16,27,30,33]
+            self.contour_pts = [0,2,14,16,27,30,33]
         self.reset()
         self.alive = False
         self.coord = None
@@ -767,9 +767,13 @@ def estimate_depth(self, face_info):
         if not face_info.rotation is None:
             success, face_info.rotation, face_info.translation = cv2.solvePnP(face_info.contour, image_pts, self.camera, self.dist_coeffs, useExtrinsicGuess=True, rvec=np.transpose(face_info.rotation), tvec=np.transpose(face_info.translation), flags=cv2.SOLVEPNP_ITERATIVE)
         else:
+            # Include jaw point for initial estimate to increase stability
+            image_pts = np.array(lms)[face_info.contour_pts + [8], 0:2]
+            contour = np.array(face_info.face_3d[face_info.contour_pts + [8]])
+
             rvec = np.array([0, 0, 0], np.float32)
             tvec = np.array([0, 0, 0], np.float32)
-            success, face_info.rotation, face_info.translation = cv2.solvePnP(face_info.contour, image_pts, self.camera, self.dist_coeffs, useExtrinsicGuess=True, rvec=rvec, tvec=tvec, flags=cv2.SOLVEPNP_ITERATIVE)
+            success, face_info.rotation, face_info.translation = cv2.solvePnP(contour, image_pts, self.camera, self.dist_coeffs, useExtrinsicGuess=True, rvec=rvec, tvec=tvec, flags=cv2.SOLVEPNP_ITERATIVE)
 
         rotation = face_info.rotation
         translation = face_info.translation