def test_mobilenet():
detector02 = Detector(face_model="RetinaFace",
emotion_model=None,
landmark_model="MobileNet")
bboxes = detector02.detect_faces(img01)
landmarks = detector02.detect_landmarks(img01, bboxes)
assert landmarks[0].shape == (68, 2)
assert (np.any(landmarks[0][:, 0] > 0) and np.any(landmarks[0][:, 0] < w)
and np.any(landmarks[0][:, 1] > 0)
and np.any(landmarks[0][:, 1] < h))
def test_pfld():
detector03 = Detector(face_model="RetinaFace",
emotion_model=None,
landmark_model="PFLD")
bboxes = detector03.detect_faces(img01)
landmarks = detector03.detect_landmarks(img01, bboxes)
assert landmarks[0].shape == (68, 2)
assert (np.any(landmarks[0][:, 0] > 0) and np.any(landmarks[0][:, 0] < w)
and np.any(landmarks[0][:, 1] > 0)
and np.any(landmarks[0][:, 1] < h))
def test_jaanet():
# AU Detection Case:
detector1 = Detector(
face_model="RetinaFace",
emotion_model=None,
landmark_model="MobileFaceNet",
au_model="jaanet",
)
bboxes = detector1.detect_faces(img01)
lands = detector1.detect_landmarks(img01, bboxes)
aus = detector1.detect_aus(img01, lands)
assert np.sum(np.isnan(aus)) == 0
assert aus.shape[-1] == 12
def test_pnp():
# Test that facepose can be estimated properly using landmarks + pnp algorithm
detector = Detector(face_model="RetinaFace",
landmark_model="MobileFaceNet",
facepose_model="PnP")
bboxes = detector.detect_faces(frame=img01)
lms = detector.detect_landmarks(frame=img01, detected_faces=bboxes)
poses = detector.detect_facepose(frame=img01, landmarks=lms)
pose_to_test = poses[0][0] # first image and first face
pitch, roll, yaw = pose_to_test.reshape(-1)
assert -10 < pitch < 10
assert -5 < roll < 5
assert -10 < yaw < 10
def test_drml():
# AU Detection Case2:
inputFname = os.path.join(get_test_data_path(), "sampler0000.jpg")
img01 = cv2.imread(inputFname)
detector1 = Detector(
face_model="RetinaFace",
emotion_model=None,
landmark_model="MobileFaceNet",
au_model="drml",
)
bboxes = detector1.detect_faces(img01)
lands = detector1.detect_landmarks(img01, bboxes)
aus = detector1.detect_aus(img01, lands)
assert np.sum(np.isnan(aus)) == 0
assert aus.shape[-1] == 12
def test_mtcnn():
detector03 = Detector(
face_model="MTCNN",
landmark_model=None,
au_model=None,
emotion_model=None,
n_jobs=1,
)
out = detector03.detect_faces(img01)
bbox_x = out[0][0]
assert bbox_x != None
bbox_width = out[0][1]
bbox_y = out[0][2]
bbox_height = out[0][3]
assert len(out[0]) == 5
assert bbox_x > 180 and bbox_x < 200
def test_retinaface():
detector02 = Detector(
face_model="RetinaFace",
landmark_model=None,
au_model=None,
emotion_model=None,
n_jobs=1,
)
out = detector02.detect_faces(img01)
bbox_x = out[0][0]
assert bbox_x != None
bbox_width = out[0][1]
bbox_y = out[0][2]
bbox_height = out[0][3]
assert len(out[0]) == 5
assert bbox_x > 180 and bbox_x < 200
def test_faceboxes():
# Face Detector Test Case:
detector01 = Detector(
face_model="FaceBoxes",
landmark_model=None,
au_model=None,
emotion_model=None,
n_jobs=1,
)
out = detector01.detect_faces(img01)
bbox_x = out[0][0]
assert bbox_x != None
bbox_width = out[0][1]
bbox_y = out[0][2]
bbox_height = out[0][3]
assert len(out[0]) == 5
assert bbox_x > 180 and bbox_x < 200
def test_img2pose():
# Test that both face detection and facepose estimation work
detector = Detector(face_model="img2pose", facepose_model="img2pose")
# Face detection
faces = detector.detect_faces(img01)[0]
bbox_x = faces[0][0]
assert bbox_x is not None
assert len(faces[0]) == 5
assert 180 < bbox_x < 200
# Pose estimation
poses = detector.detect_facepose(img01)[0]
pose_to_test = poses[0][0] # first image and first face
pitch, roll, yaw = pose_to_test.reshape(-1)
assert -10 < pitch < 10
assert -5 < roll < 5
assert -10 < yaw < 10
def test_svm():
# AU Detection Case:
detector1 = Detector(
face_model="RetinaFace",
emotion_model=None,
landmark_model="MobileFaceNet",
au_model="svm",
)
detected_faces = detector1.detect_faces(img01)
landmarks = detector1.detect_landmarks(img01, detected_faces)
hogs, new_lands = detector1._batch_hog(frames=img01,
detected_faces=detected_faces,
landmarks=landmarks)
aus = detector1.detect_aus(frame=hogs, landmarks=new_lands)
assert np.sum(np.isnan(aus)) == 0
assert aus.shape[-1] == 20
def test_mtcnn():
detector03 = Detector(
face_model="MTCNN",
landmark_model=None,
au_model=None,
emotion_model=None,
n_jobs=1,
)
out = detector03.detect_faces(img01)
bbox_left = out[0][0]
assert bbox_left != None
bbox_right = out[0][1]
bbox_top = out[0][2]
bbox_bottom = out[0][3]
assert len(out[0]) == 5
assert (bbox_left > 0 and bbox_right > 0 and bbox_top > 0
and bbox_bottom > 0 and bbox_left < bbox_right
and bbox_top < bbox_bottom and bbox_left < w and bbox_right < w
and bbox_top < h and bbox_bottom < h)
def test_rf():
# AU Detection Case:
detector1 = Detector(
face_model="RetinaFace",
emotion_model=None,
landmark_model="MobileFaceNet",
au_model="RF",
)
bboxes = detector1.detect_faces(img01)
lands = detector1.detect_landmarks(img01, bboxes)
convex_hull, new_lands = detector1.extract_face(
frame=img01,
detected_faces=[bboxes[0:4]],
landmarks=lands,
size_output=112)
hogs = detector1.extract_hog(frame=convex_hull, visualize=False)
aus = detector1.detect_aus(frame=hogs, landmarks=new_lands)
assert np.sum(np.isnan(aus)) == 0
assert aus.shape[-1] == 20
def test_faceboxes():
# Face Detector Test Case:
detector01 = Detector(
face_model="FaceBoxes",
landmark_model=None,
au_model=None,
emotion_model=None,
n_jobs=1,
)
out = detector01.detect_faces(img01)
bbox_left = out[0][0]
assert bbox_left != None
bbox_right = out[0][1]
bbox_top = out[0][2]
bbox_bottom = out[0][3]
assert len(out[0]) == 5
assert (bbox_left > 0 and bbox_right > 0 and bbox_top > 0
and bbox_bottom > 0 and bbox_left < bbox_right
and bbox_top < bbox_bottom and bbox_left < w and bbox_right < w
and bbox_top < h and bbox_bottom < h)