Merge pull request #85 from serengil/feat-task-2502-align-first-by-default

align first by default

Author: serengil

.github/pull_request_template.md

@@ -0,0 +1,13 @@
+## Tickets
+
+https://github.com/serengil/retinaface/issues/XXX
+
+### What has been done
+
+With this PR, ...
+
+## How to test
+
+```shell
+make lint && make test
+```

retinaface/RetinaFace.py

@@ -212,7 +212,6 @@ def extract_faces(
     align: bool = True,
     allow_upscaling: bool = True,
     expand_face_area: int = 0,
-    align_first: bool = False,
 ) -> list:
     """
     Extract detected and aligned faces
@@ -223,8 +222,6 @@ def extract_faces(
         align (bool): enable or disable alignment
         allow_upscaling (bool): allowing up-scaling
         expand_face_area (int): expand detected facial area with a percentage
-        align_first (bool): set this True to align first and detect second
-            this can be applied only if input image has just one face
     """
     resp = []
 
@@ -238,12 +235,6 @@ def extract_faces(
         img_path=img, threshold=threshold, model=model, allow_upscaling=allow_upscaling
     )
 
-    if align_first is True and len(obj) > 1:
-        logger.warn(
-            f"Even though align_first is set to True, there are {len(obj)} faces in input image."
-            "Align first functionality can be applied only if there is single face in the input"
-        )
-
     if not isinstance(obj, dict):
         return resp
 
@@ -263,10 +254,7 @@ def extract_faces(
         x2 = min(img.shape[1], w + int((w * expand_face_area) / 100))  # expand right
         y2 = min(img.shape[0], h + int((h * expand_face_area) / 100))  # expand bottom
 
-        if align_first is False or (align_first is True and len(obj) > 1):
-            facial_img = img[y1:y2, x1:x2]
-        else:
-            facial_img = img.copy()
+        facial_img = img[y1:y2, x1:x2]
 
         if align is True:
             landmarks = identity["landmarks"]
@@ -275,19 +263,33 @@ def extract_faces(
             nose = landmarks["nose"]
             # mouth_right = landmarks["mouth_right"]
             # mouth_left = landmarks["mouth_left"]
+
+            # notice that left eye of one is seen on the right from your perspective
             facial_img, rotate_angle, rotate_direction = postprocess.alignment_procedure(
-                facial_img, right_eye, left_eye, nose
+                img=img, left_eye=right_eye, right_eye=left_eye, nose=nose
             )
 
-        if align_first is True and len(obj) == 1:
-            facial_area = postprocess.rotate_facial_area(
+            # find new facial area coordinates after alignment
+            projected_facial_area = postprocess.rotate_facial_area(
                 facial_area, rotate_angle, rotate_direction, img.shape
             )
             # Expand the facial area to be extracted and stay within img.shape limits
-            x1 = max(0, facial_area[0] - int((facial_area[2] * expand_face_area) / 100))
-            y1 = max(0, facial_area[1] - int((facial_area[3] * expand_face_area) / 100))
-            x2 = min(img.shape[1], facial_area[2] + int((facial_area[2] * expand_face_area) / 100))
-            y2 = min(img.shape[0], facial_area[3] + int((facial_area[3] * expand_face_area) / 100))
+            x1 = max(
+                0,
+                projected_facial_area[0] - int((projected_facial_area[2] * expand_face_area) / 100),
+            )
+            y1 = max(
+                0,
+                projected_facial_area[1] - int((projected_facial_area[3] * expand_face_area) / 100),
+            )
+            x2 = min(
+                img.shape[1],
+                projected_facial_area[2] + int((projected_facial_area[2] * expand_face_area) / 100),
+            )
+            y2 = min(
+                img.shape[0],
+                projected_facial_area[3] + int((projected_facial_area[3] * expand_face_area) / 100),
+            )
             facial_img = facial_img[y1:y2, x1:x2]
 
         resp.append(facial_img[:, :, ::-1])

tests/test_actions.py

@@ -75,7 +75,7 @@ def test_alignment_for_clock_way():
 
 
 def do_alignment_checks(img: np.ndarray, expected_faces: int) -> None:
-    faces = RetinaFace.extract_faces(img_path=img, align=True, expand_face_area=100)
+    faces = RetinaFace.extract_faces(img_path=img, align=True, expand_face_area=10)
 
     # it has one clear face
     assert len(faces) == expected_faces

tests/test_align_first.py

@@ -6,24 +6,15 @@
 
 THRESHOLD = 1000
 
-
-def test_detect_first():
-    """
-    Test the default behavior. Detect first and align second causes
-        so many black pixels
-    """
-    faces = RetinaFace.extract_faces(img_path="tests/dataset/img11.jpg")
-    num_black_pixels = np.sum(np.all(faces[0] == 0, axis=2))
-    assert num_black_pixels > THRESHOLD
-    logger.info("✅ Disabled align_first test for single face photo done")
+VISUAL_TEST = False
 
 
 def test_align_first():
     """
     Test align first behavior. Align first and detect second do not cause
         so many black pixels in contrast to default behavior
     """
-    faces = RetinaFace.extract_faces(img_path="tests/dataset/img11.jpg", align_first=True)
+    faces = RetinaFace.extract_faces(img_path="tests/dataset/img11.jpg")
     num_black_pixels = np.sum(np.all(faces[0] == 0, axis=2))
     assert num_black_pixels < THRESHOLD
     logger.info("✅ Enabled align_first test for single face photo  done")
@@ -34,22 +25,14 @@ def test_align_first_for_group_photo():
     Align first will not work if the given image has many faces and
         it will cause so many black pixels
     """
-    faces = RetinaFace.extract_faces(img_path="tests/dataset/couple.jpg", align_first=True)
-    for face in faces:
-        num_black_pixels = np.sum(np.all(face == 0, axis=2))
-        assert num_black_pixels > THRESHOLD
-
-    logger.info("✅ Enabled align_first test for group photo done")
-
-
-def test_default_behavior_for_group_photo():
-    """
-    Align first will not work in the default behaviour and
-        it will cause so many black pixels
-    """
     faces = RetinaFace.extract_faces(img_path="tests/dataset/couple.jpg")
     for face in faces:
         num_black_pixels = np.sum(np.all(face == 0, axis=2))
-        assert num_black_pixels > THRESHOLD
+        assert num_black_pixels < THRESHOLD
+        if VISUAL_TEST is True:
+            import matplotlib.pyplot as plt
 
-    logger.info("✅ Disabled align_first test for group photo done")
+            plt.imshow(face)
+            plt.show()
+
+    logger.info("✅ Enabled align_first test for group photo done")