def train_arcface_model(face_detector, config, dataset_folder,
embedding_folder):
objs = []
register_image_bbox_objs = []
registered_ids = []
for folder in os.listdir(dataset_folder):
if not folder.endswith('.txt'):
for face in os.listdir(f"{dataset_folder}/{folder}"):
raw_image_path = f"{dataset_folder}/{folder}/{face}"
train_image_id = ImageId(channel='train_img',
timestamp=arrow.now().timestamp,
file_format=os.path.splitext(face)[1])
train_image_obj = Image(train_image_id,
raw_image_path=raw_image_path)
try:
face_detection_result = face_detector.detect(
train_image_obj, label=folder)
if len(face_detection_result.detected_objects) == 1:
register_image_bbox_objs.append(
face_detection_result.detected_objects[0])
registered_ids.append(
face_detection_result.detected_objects[0].label)
objs += train_image_obj.fetch_bbox_pil_objs(
register_image_bbox_objs)
except:
print(
f"[ERROR] An error occured with the face detector model. Can't open the photo {dataset_folder}/{folder}/{face}"
)
register_image_bbox_objs = []
objects_frame = resize_and_stack_image_objs((112, 112), objs)
print("Object_frame shape:", objects_frame.shape)
objects_frame = np.transpose(objects_frame, (0, 3, 1, 2))
with SimpleTimer("[INFO] Extracting embedding for our dataset"):
arcface_classifier = ArcFaceClassifier(config,
registered_ids,
objects_frame=objects_frame)
embedding_path = f"{embedding_folder}/faces.pkl"
print(f"[INFO] Store face embedding to {embedding_path}")
arcface_classifier.store_embedding_info(embedding_path)
ids_path = f"{embedding_folder}/registered_ids.txt"
print(f"[INFO] Store registered ids to {ids_path}")
with open(ids_path, 'w') as filehandle:
for registered_id in registered_ids:
filehandle.write('%s\n' % registered_id)
return arcface_classifier
y2=1007,
label='Zoe',
score=1,
meta=''),
#BoundedBoxObject(x1=946, y1=551, x2=1149, y2=784, label='Amine', score=1, meta=''),
BoundedBoxObject(x1=1364,
y1=564,
x2=1492,
y2=720,
label='Lu',
score=1,
meta='')
#BoundedBoxObject(x1=2286, y1=1272, x2=2629, y2=1738, label='Hedi', score=1, meta='')
]
objs = train_image_obj.fetch_bbox_pil_objs(register_image_bbox_objs)
objects_frame = resize_and_stack_image_objs((112, 112), objs)
objects_frame = np.transpose(objects_frame, (0, 3, 1, 2))
registered_ids = [i.label for i in register_image_bbox_objs]
# Création du classifier et réentrainement
model_name = 'Arcface'
with SimpleTimer("Loading model %s" % model_name):
arcface_classifier = ArcFaceClassifier(registered_ids,
objects_frame=objects_frame)
# Test de reconnaissance
raw_image_path = 'demo/183club/test_imag2.jpg'
test_image_id = ImageId(channel='demo',
timestamp=arrow.now().timestamp,
file_format='jpg')