Implement get_landmarks

pyproject.toml

@@ -11,6 +11,7 @@ requires-python = "~3.13"
 dependencies = [
     "numpy (>=2.3.3,<3.0.0)",
     "face-alignment (>=1.4.1,<2.0.0)",
+    "scikit-image (>=0.25.2,<0.26.0)",
 ]
 
 
@@ -24,3 +25,4 @@ torch = [
     {url = "https://download.pytorch.org/whl/cu129/torch-2.8.0%2Bcu129-cp313-cp313-win_amd64.whl", platform = "win32"},
 ]
 pytest = "^8.4.2"
+matplotlib = "^3.10.6"

src/falign/landmarks.py

@@ -0,0 +1,21 @@
+from numpy.typing import NDArray
+
+from pathlib import Path
+
+from skimage.io import imread
+import torch
+import face_alignment
+
+
+def _get_face_landmarks(image: NDArray, face_aligner: face_alignment.FaceAlignment):
+    landmarks = face_aligner.get_landmarks(image)
+    assert landmarks is not None, "No face detected"
+
+    return landmarks[0]
+
+
+def get_landmarks(path_image: Path, dtype: torch.dtype = torch.float32, device: torch.device = torch.device("cuda")):
+    face_aligner = face_alignment.FaceAlignment(face_alignment.LandmarksType.TWO_D, dtype=dtype, device=device.type)
+    image = imread(path_image)
+
+    return _get_face_landmarks(image, face_aligner)

test/test_falign/test_landmarks.py

@@ -0,0 +1,22 @@
+from pathlib import Path
+
+from matplotlib import pyplot
+
+from falign.landmarks import get_landmarks
+
+
+def test_get_landmarks():
+    dir_gallery = Path(__file__).parent.parent.parent / "gallery"
+    path_image = dir_gallery / "original.jpg"
+    landmarks = get_landmarks(path_image)
+
+    assert landmarks.shape == (68, 2)
+
+    fig, ax = pyplot.subplots()
+    ax.imshow(pyplot.imread(path_image))
+    ax.scatter(landmarks[:, 0], landmarks[:, 1], s=1)
+    ax.axis("off")
+    ax.margins(0)
+    ax.set_aspect("equal")
+    fig.tight_layout(pad=0)
+    fig.savefig(dir_gallery / "test_get_landmarks.jpg")