def infer_object_detections_on_loaded_image(
self,
image_np: np.array,
):
orig_img_height, orig_img_width = image_np.shape[:2]
img_np = resize_and_letter_box(image_np / 255.,
target_width=self.model_image_width,
target_height=self.model_image_height)
img_np = np.expand_dims(img_np, 0)
start = time.time()
detected_boxes, detected_classes, detected_scores = infer_objects_in_image(
image=img_np,
session=self.session,
orig_image_height=orig_img_height,
orig_image_width=orig_img_width,
out_tensors=self.out_tensors,
input_tensor=self.input_tensors[0],
orig_image_width_placeholder_tensor=self.input_tensors[1],
orig_image_height_placeholder_tensor=self.input_tensors[2])
end = time.time()
time_in_s = end - start
if self.verbose:
print(f'Took {time_in_s} seconds to run prediction in tf session.')
return detected_boxes, detected_classes, detected_scores
def infer_object_detections_on_loaded_image(
self,
image_np: np.array,
):
orig_img_height, orig_img_width = image_np.shape[:2]
img_np = resize_and_letter_box(image_np / 255.,
target_width=608,
target_height=608)
img_np = np.expand_dims(img_np, 0)
detected_boxes, detected_classes, detected_scores = infer_objects_in_image(
image=img_np * 255.,
model=self.model,
orig_image_height=orig_img_height,
orig_image_width=orig_img_width,
detection_prob_treshold=self.detection_prob_treshold)
return detected_boxes, detected_classes, detected_scores
def infer_object_detections_on_loaded_image(
self,
image_np: np.array,
):
"""
Infers object detection on the loaded numpy array
representing pixels of the image (row-major order).
:param image_np np.array containing pixels of the image (row-major ordering)
:return (detected_boxes, detected_classes, detected_scores)
- detected_boxes array of (left, top, bottom, right)
- detected_classes array of ints representing class indices
- detected_scores array of floats representing probability for each box and class
"""
orig_img_height, orig_img_width = image_np.shape[:2]
img_np = resize_and_letter_box(
image_np / 255.,
target_width=self.model_image_width,
target_height=self.model_image_height,
interpolation=self.interpolation_strategy)
img_np = np.expand_dims(img_np, 0)
start = time.time()
detected_boxes, detected_classes, detected_scores = infer_objects_in_image(
image=img_np,
session=self.session,
orig_image_height=orig_img_height,
orig_image_width=orig_img_width,
out_tensors=self.out_tensors,
input_tensor=self.input_tensors[0],
orig_image_width_placeholder_tensor=self.input_tensors[1],
orig_image_height_placeholder_tensor=self.input_tensors[2])
end = time.time()
time_in_s = end - start
if self.verbose:
print(f'Took {time_in_s} seconds to run prediction in tf session.')
return detected_boxes, detected_classes, detected_scores
def infer_object_detections_on_loaded_image(
self,
image_np: np.array,
):
orig_img_height, orig_img_width = image_np.shape[:2]
img_np = resize_and_letter_box(image_np / 255.,
target_width=self.model_image_width,
target_height=self.model_image_height)
img_np = np.expand_dims(img_np, 0)
detected_boxes, detected_classes, detected_scores = infer_objects_in_image(
image=img_np,
restored_model=self.model,
session=self.session,
orig_image_height=orig_img_height,
orig_image_width=orig_img_width,
detection_prob_treshold=self.detection_threshold,
model_image_height=self.model_image_height,
model_image_width=self.model_image_width,
anchors=self.anchors,
verbose=self.verbose)
return detected_boxes, detected_classes, detected_scores