def __init__(
self,
model_spec: Union[YOLOv5_ModelSpec, YOLOv5_TFLite_ModelSpec]
):
super().__init__(model_spec)
if model_spec.class_names is not None:
if isinstance(model_spec.class_names, str) or isinstance(model_spec.class_names, Path):
with fsspec.open(model_spec.class_names, 'r', encoding='utf-8') as out:
self.class_names = np.array(json.load(out))
else:
self.class_names = np.array(model_spec.class_names)
else:
self.class_names = None
if isinstance(model_spec.preprocess_input, str) or isinstance(model_spec.preprocess_input, Path):
self._preprocess_input = get_preprocess_input_from_script_file(
script_file=model_spec.preprocess_input
)
else:
if model_spec.preprocess_input is None:
self._preprocess_input = lambda x: x
else:
self._preprocess_input = model_spec.preprocess_input
if isinstance(model_spec, YOLOv5_ModelSpec):
self._load_yolov5_model(model_spec)
self._raw_predict_images = self._raw_predict_images_torch
elif isinstance(model_spec, YOLOv5_TFLite_ModelSpec):
self._load_yolov5_tflite(model_spec)
self._raw_predict_images = self._raw_predict_images_tflite
else:
raise ValueError(
f"ObjectDetectionAPI_Model got unknown DetectionModelSpec: {type(model_spec)}"
)
def __init__(
self,
model_spec: Union[
ObjectDetectionAPI_ModelSpec,
ObjectDetectionAPI_pb_ModelSpec,
ObjectDetectionAPI_TFLite_ModelSpec,
ObjectDetectionAPI_KFServing
],
):
super().__init__(model_spec)
if model_spec.class_names is not None:
if isinstance(model_spec.class_names, str) or isinstance(model_spec.class_names, Path):
with fsspec.open(model_spec.class_names, 'r', encoding='utf-8') as out:
self.class_names = np.array(json.load(out))
else:
self.class_names = np.array(model_spec.class_names)
if isinstance(model_spec, ObjectDetectionAPI_pb_ModelSpec):
self.class_names_coef = -1 # saved_model.pb returns from 1
else:
self.class_names_coef = 0
else:
self.class_names = None
self.coef = -1
if isinstance(model_spec, ObjectDetectionAPI_ModelSpec):
self._load_object_detection_api(model_spec)
self._raw_predict_single_image = self._raw_predict_single_image_default
elif isinstance(model_spec, ObjectDetectionAPI_pb_ModelSpec):
self._load_object_detection_api_pb(model_spec)
self._raw_predict_single_image = self._raw_predict_single_image_default
elif isinstance(model_spec, ObjectDetectionAPI_TFLite_ModelSpec):
self._load_object_detection_api_tflite(model_spec)
self._raw_predict_single_image = self._raw_predict_single_image_tflite
elif isinstance(model_spec, ObjectDetectionAPI_KFServing):
self.input_dtype = INPUT_TYPE_TO_DTYPE[model_spec.input_type]
# Wake up the service
try:
self._raw_predict_single_image_kfserving(
image=np.zeros((128, 128, 3)),
timeout=1.
)
except requests.exceptions.ReadTimeout:
pass
self._raw_predict_single_image = self._raw_predict_single_image_kfserving
else:
raise ValueError(
f"ObjectDetectionAPI_Model got unknown DetectionModelSpec: {type(model_spec)}"
)
if isinstance(model_spec.preprocess_input, str) or isinstance(model_spec.preprocess_input, Path):
self._preprocess_input = get_preprocess_input_from_script_file(
script_file=model_spec.preprocess_input
)
else:
if model_spec.preprocess_input is None:
self._preprocess_input = lambda x: x
else:
self._preprocess_input = model_spec.preprocess_input
def __init__(
self,
model_spec: Union[TensorFlow_KeypointsRegressorModelSpec,
TensorFlow_KeypointsRegressorModelSpec_TFServing]):
super().__init__(model_spec)
if isinstance(model_spec, TensorFlow_KeypointsRegressorModelSpec):
self._load_tensorflow_KeypointsRegressor_model_spec(model_spec)
self._raw_predict = self._raw_predict_tensorflow
elif isinstance(model_spec,
TensorFlow_KeypointsRegressorModelSpec_TFServing):
self.input_dtype = INPUT_TYPE_TO_DTYPE[model_spec.input_type]
# Wake up the service
try:
self._raw_predict_kfserving(images=np.zeros(
(1, *self.input_size, 3)),
timeout=1.)
except requests.exceptions.ReadTimeout:
pass
self._raw_predict = self._raw_predict_kfserving
else:
raise ValueError(
f"Tensorflow_KeypointsRegressorModel got unknown KeypointsRegressorModelSpec: {type(model_spec)}"
)
if isinstance(model_spec.preprocess_input, str) or isinstance(
model_spec.preprocess_input, Path):
self._preprocess_input = get_preprocess_input_from_script_file(
script_file=model_spec.preprocess_input)
else:
if model_spec.preprocess_input is None:
self._preprocess_input = lambda x: x
else:
self._preprocess_input = model_spec.preprocess_input
def __init__(self, model_spec: Detectron2_ModelSpec):
super().__init__(model_spec)
if model_spec.class_names is not None:
if isinstance(model_spec.class_names, str) or isinstance(
model_spec.class_names, Path):
with fsspec.open(model_spec.class_names, 'r',
encoding='utf-8') as out:
self.class_names = np.array(json.load(out))
else:
self.class_names = np.array(model_spec.class_names)
else:
self.class_names = None
if isinstance(model_spec, Detectron2_ModelSpec):
self._load_pt_model(model_spec)
self.device = model_spec.device
self.input_format = model_spec.device
self.input_type = model_spec.input_type
else:
raise ValueError(
f"{Detectron2_DetectionModel.__name__} got unknown DetectionModelSpec: {type(model_spec)}"
)
if isinstance(model_spec.preprocess_input, str) or isinstance(
model_spec.preprocess_input, Path):
self._preprocess_input = get_preprocess_input_from_script_file(
script_file=model_spec.preprocess_input)
else:
if model_spec.preprocess_input is None:
self._preprocess_input = lambda x: x
else:
self._preprocess_input = model_spec.preprocess_input
def __init__(
self,
model_spec: PyTorch_EmbedderModelSpec
):
super().__init__(model_spec)
self._load_pytorch_model_spec(model_spec)
self._raw_predict = self._raw_predict_pytorch
if isinstance(model_spec.preprocess_input, str) or isinstance(model_spec.preprocess_input, Path):
self._preprocess_input = get_preprocess_input_from_script_file(
script_file=model_spec.preprocess_input
)
else:
if model_spec.preprocess_input is None:
self._preprocess_input = lambda x: x
else:
self._preprocess_input = model_spec.preprocess_input
def __init__(
self, model_spec: Union[TensorFlow_ClassificationModelSpec,
TensorFlow_ClassificationModelSpec_TFServing]):
super().__init__(model_spec)
if isinstance(model_spec.class_names, str) or isinstance(
model_spec.class_names, Path):
with fsspec.open(model_spec.class_names, 'r',
encoding='utf-8') as out:
self._class_names = json.load(out)
else:
self._class_names = model_spec.class_names
if isinstance(model_spec, TensorFlow_ClassificationModelSpec):
self._load_tensorflow_classification_model_spec(model_spec)
self._raw_predict = self._raw_predict_tensorflow
elif isinstance(model_spec,
TensorFlow_ClassificationModelSpec_TFServing):
self.input_dtype = INPUT_TYPE_TO_DTYPE[model_spec.input_type]
# Wake up the service
try:
self._raw_predict_kfserving(images=np.zeros(
(1, *self.input_size, 3)),
timeout=1.)
except requests.exceptions.ReadTimeout:
pass
self._raw_predict = self._raw_predict_kfserving
else:
raise ValueError(
f"Tensorflow_ClassificationModel got unknown ClassificationModelSpec: {type(model_spec)}"
)
if isinstance(model_spec.preprocess_input, str) or isinstance(
model_spec.preprocess_input, Path):
self._preprocess_input = get_preprocess_input_from_script_file(
script_file=model_spec.preprocess_input)
else:
if model_spec.preprocess_input is None:
self._preprocess_input = lambda x: x
else:
self._preprocess_input = model_spec.preprocess_input
self.id_to_class_name = np.array(
[class_name for class_name in self._class_names])