version_upgrade_for_mac

.github/workflows/flake8.yml

@@ -11,7 +11,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: ["3.7"]
+        python-version: ["3.8"]
     steps:
     - uses: actions/checkout@v3
     - name: Set up Python ${{ matrix.python-version }}
@@ -37,7 +37,7 @@ jobs:
     - name: Set up Python
       uses: actions/setup-python@v2
       with:
-        python-version: 3.7
+        python-version: 3.8
 
     - name: Install dependencies
       run: |
@@ -57,7 +57,7 @@ jobs:
     - name: Set up Python
       uses: actions/setup-python@v2
       with:
-        python-version: 3.7
+        python-version: 3.8
 
     - name: Install dependencies
       run: |

.github/workflows/pytest.yml

@@ -22,7 +22,7 @@ jobs:
         with:
           activate-environment: l2d-vtb
           environment-file: environment.yml
-          python-version: "3.7"
+          python-version: "3.8"
       - run: |
           conda env list
           conda activate l2d-vtb
@@ -51,7 +51,7 @@ jobs:
         with:
           activate-environment: l2d-vtb
           environment-file: environment.yml
-          python-version: "3.7"
+          python-version: "3.8"
       - run: |
           conda env list
           conda activate l2d-vtb
@@ -61,7 +61,7 @@ jobs:
           pip install -r requirements.txt
           conda install -c conda-forge dlib
           pip install scipy>0.16
-          conda install pytorch==1.10.2 torchvision==0.11.3 torchaudio==0.10.2 -c pytorch
+          conda install pytorch torchvision torchaudio -c pytorch-nightly
     
       - name: Run tests
         run: |
@@ -80,7 +80,7 @@ jobs:
         with:
           activate-environment: l2d-vtb
           environment-file: environment.yml
-          python-version: "3.7"
+          python-version: "3.8"
       - run: |
           conda env list
           conda activate l2d-vtb

environment.yml

@@ -3,7 +3,7 @@ channels:
   - defaults
   - conda-forge
 dependencies:
-  - python=3.7
+  - python=3.8
   - pip
   - flake8
   - pytest

face_alignment/api.py

@@ -60,6 +60,9 @@ def __init__(self, landmarks_type, network_size=NetworkSize.LARGE, use_onnx=Fals
 
         if 'cuda' in device:
             torch.backends.cudnn.benchmark = True
+
+        if 'mps' in device:
+            torch.backends.mps.benchmark = True   
 
         # Get the face detector
         face_detector_module = __import__('face_alignment.detection.' + face_detector,

face_alignment/detection/core.py

@@ -25,9 +25,9 @@ def __init__(self, device, verbose):
                 logger = logging.getLogger(__name__)
                 logger.warning("Detection running on CPU, this may be potentially slow.")
 
-        if 'cpu' not in device and 'cuda' not in device:
+        if 'cpu' not in device and 'cuda' not in device and 'mps' not in device:
             if verbose:
-                logger.error("Expected values for device are: {cpu, cuda} but got: %s", device)
+                logger.error("Expected values for device are: {cpu, cuda, mps} but got: %s", device)
             raise ValueError
 
     def detect_from_image(self, tensor_or_path):
@@ -123,7 +123,7 @@ def tensor_or_path_to_ndarray(tensor_or_path, rgb=True):
         if isinstance(tensor_or_path, str):
             return cv2.imread(tensor_or_path) if not rgb else io.imread(tensor_or_path)
         elif torch.is_tensor(tensor_or_path):
-            # Call cpu in case its coming from cuda
+            # Call cpu in case its coming from cuda or mps
             return tensor_or_path.cpu().numpy()[..., ::-1].copy() if not rgb else tensor_or_path.cpu().numpy()
         elif isinstance(tensor_or_path, np.ndarray):
             return tensor_or_path[..., ::-1].copy() if not rgb else tensor_or_path

face_alignment/detection/dlib/dlib_detector.py

@@ -18,7 +18,7 @@ def __init__(self, device, path_to_detector=None, verbose=False):
         base_path = os.path.join(appdata_dir('face_alignment'), "data")
 
         # Initialise the face detector
-        if 'cuda' in device:
+        if 'cuda' in device or 'mps' in device:
             if path_to_detector is None:
                 path_to_detector = os.path.join(
                     base_path, "mmod_human_face_detector.dat")
@@ -47,7 +47,7 @@ def detect_from_image(self, tensor_or_path):
 
         detected_faces = self.face_detector(cv2.cvtColor(image, cv2.COLOR_BGR2GRAY))
 
-        if 'cuda' not in self.device:
+        if 'cuda' not in self.device and 'mps' not in self.device:
             detected_faces = [[d.left(), d.top(), d.right(), d.bottom()] for d in detected_faces]
         else:
             detected_faces = [[d.rect.left(), d.rect.top(), d.rect.right(), d.rect.bottom()] for d in detected_faces]

face_alignment/detection/sfd/detect.py

@@ -12,6 +12,9 @@ def detect(net, img, device):
 
     if 'cuda' in device:
         torch.backends.cudnn.benchmark = True
+
+    if 'mps' in device:
+        torch.backends.mps.benchmark = True
 
     img = torch.FloatTensor(img).to(device)
     BB, CC, HH, WW = img.size()

face_pose/pose_estimator.py

@@ -134,7 +134,7 @@ def draw_annotation_box(self, image, rotation_vector, translation_vector, color=
         point_3d.append((front_size, front_size, front_depth))
         point_3d.append((front_size, -front_size, front_depth))
         point_3d.append((-front_size, -front_size, front_depth))
-        point_3d = np.array(point_3d, dtype=np.float).reshape(-1, 3)
+        point_3d = np.array(point_3d, dtype=np.float32).reshape(-1, 3)
 
         # Map to 2d image points
         (point_2d, _) = cv2.projectPoints(point_3d,

main.py

@@ -2,8 +2,6 @@
 import cv2
 import numpy as np
 import time
-import dlib
-import face_alignment
 import utils
 
 from argparse import ArgumentParser
@@ -16,20 +14,34 @@
 
 
 def run():
-    # Define operating system
-    cap = utils.def_cap(args.cam)
+    # Get operating system
+    os = utils.get_os()
+
+    if os == 'Windows':  # Windows OS
+        cap = cv2.VideoCapture(args.cam + cv2.CAP_DSHOW)
+    else:  # Linux & Mac OS
+        cap = cv2.VideoCapture(args.cam)
+
     cap.set(cv2.CAP_PROP_FPS, 30)
     cap.set(cv2.CAP_PROP_FRAME_WIDTH, 640)
     _, sample_frame = cap.read()
 
     # Setup face detection models
     if not args.gpu:  # CPU: use dlib
+        import dlib
         dlib_model_path = 'face_pose/shape_predictor_68_face_landmarks.dat'
         shape_predictor = dlib.shape_predictor(dlib_model_path)
         face_detector = dlib.get_frontal_face_detector()
+
     else:  # GPU: use FAN (better)
-        fa = face_alignment.FaceAlignment(
-            face_alignment.LandmarksType._2D, flip_input=False)
+        import face_alignment
+        if os == 'Linux':  # Linux
+            fa = face_alignment.FaceAlignment(
+                face_alignment.LandmarksType._2D, device='cuda', flip_input=False)
+        if os == 'Darwin':  # MacOS
+            fa = face_alignment.FaceAlignment(
+                face_alignment.LandmarksType._2D, device='mps', flip_input=False)
+
         face_detector = fa.face_detector
 
     # Introduce pose estimator to solve pose, get one frame to setup the estimator according to the image size

requirements.txt

@@ -1,3 +1,3 @@
-numpy==1.21.6
-numba==0.56.3
-opencv_python==4.4.0.44
+numpy==1.24.4
+numba==0.58.1
+opencv_python==4.8.1.78

tests/test_mac.py

@@ -45,8 +45,8 @@ def test_module_install_gpu():
         v1 = pkg_resources.get_distribution(modules[1]).version
         assert v1 is not None
 
-        # Test CUDA
-        if m1.cuda.is_available():
+        # Test M1 GPU
+        if m1.backends.mps.is_available() and m1.backends.mps.is_built():
             assert True
         else:
             print(m1.__version__)

utils.py

@@ -3,13 +3,9 @@
 import cv2
 
 
-def def_cap(cam):
-    """Define capture"""
-    if system() in ['Windows']:  # Windows OS
-        cap = cv2.VideoCapture(cam + cv2.CAP_DSHOW)
-    else:  # Linux & Mac OS
-        cap = cv2.VideoCapture(cam)
-    return cap
+def get_os():
+    """Get operating system"""
+    return system()
 
 
 def get_face(detector, image, gpu=False):