def __init__(self, name, input_img_channel='rgb'):
name_model, inference, tpu = name.split('_')
tpu = (tpu == 'tpu')
assert inference in ['fp32',
'int8'], 'Available inferences are: fp32, int8'
name_model = name_model + '_' + inference
if tpu:
delegates = [load_delegate('libedgetpu.so.1')]
name_model = name_model + '_tpu'
else:
name_model = name_model + '_cpu'
delegates = []
path_tflite = helper.get_path('landmark_detection', name_model)
self.landmark_detection_inference = Interpreter(
model_path=str(path_tflite),
experimental_delegates=delegates,
)
self.landmark_detection_inference.allocate_tensors()
self.input_index = self.landmark_detection_inference.get_input_details(
)[0]['index']
self.output_index = self.landmark_detection_inference.get_output_details(
)[0]['index']
assert input_img_channel in ['bgr',
'rgb'], 'Incorrect input_img_channel'
self.input_img_channel = input_img_channel
def __init__(self, min_neighbors=5, scale=1.2, **kwargs):
name_model = 'haarcascade'
type_model = 'face_detection'
path_xml = helper.get_path(type_model, name_model)
super(HAARCascade, self).__init__(
path_xml,
min_neighbors=min_neighbors,
scale=scale,
**kwargs,
)
def __init__(self, input_img_channel='rgb', resize=None, tpu=False):
if tpu:
name_model = 'ssd_int8_tpu'
delegates = [load_delegate('libedgetpu.so.1')]
else:
name_model = 'ssd_int8_cpu'
delegates = []
path_tflite = helper.get_path('face_detection', name_model)
self.face_detection_inference = Interpreter(
model_path=str(path_tflite), experimental_delegates=delegates)
self.face_detection_inference.allocate_tensors()
self.index_input = self.face_detection_inference.get_input_details(
)[0]['index']
self.index_boxes = self.face_detection_inference.get_output_details(
)[0]['index']
self.index_probs = self.face_detection_inference.get_output_details(
)[2]['index']
assert input_img_channel in ['bgr',
'rgb'], 'Incorrect input_img_channel'
self.input_img_channel = input_img_channel