def object_detector(img):
dark_frame = pydarknet.Image(img)
results = net.detect(dark_frame)
del dark_frame
op = list()
for cat, score, bounds in results:
x, y, w, h = bounds
op.append([x, y, w, h])
return op
def yolo():
# retrieve parameters from html form
filename1 = request.form['image']
yversion = request.form['yolo_version']
# open images
target = os.path.join(APP_ROOT, 'static/images')
destination1 = "/".join([target, filename1])
img1 = Image.open(destination1)
img = np.asarray(img1)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
destination = "/".join([target, 'temp.png'])
if os.path.isfile(destination):
print("Existing file found at " + destination)
os.remove(destination)
print("Performing YOLO" + yversion + " on image at " + destination1 +
"...\nThis might take a while...")
start_time = time.time()
results = darknet[yversion].detect(pydarknet.Image(img))
end_time = time.time()
#print(results)
print("Elapsed Time:", end_time - start_time)
print("Detected:")
for cat, score, bounds in results:
x, y, w, h = bounds
label = str(cat.decode("utf-8"))
conf = float(score * 100)
displayText = str("{}, {:.2f}%".format(label, conf))
cv2.rectangle(img, (int(x - w / 2), int(y - h / 2)),
(int(x + w / 2), int(y + h / 2)), (0, 0, 255),
thickness=1)
cv2.putText(img,
displayText, (int(x - w / 2), int(y - h / 2)),
cv2.FONT_HERSHEY_SIMPLEX,
0.5, (0, 0, 255),
thickness=1,
lineType=cv2.LINE_AA)
print("> " + displayText)
# save and return image
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
im_pil = Image.fromarray(img)
im_pil.save(destination)
return send_image('temp.png')
def predict(self, image: Image) -> List[Prediction]:
"""
Predict bounding boxes for the persons in the image.
:param image: the PIL image
:return: the bounding boxes for the persons in the image.
"""
img = numpy.asarray(image)
img_darknet = pydarknet.Image(img)
detections = self._model.detect(img_darknet)
predictions = self._get_predictions_from_detections(detections)
return list(predictions)
def on_msg_camera_stream(self, msg):
self._logger.info('%s received frame %s', self.name, msg.timestamp)
start_time = time.time()
cv_img = self.bridge.imgmsg_to_cv2(msg.data, 'rgb8')
img = pydarknet.Image(cv_img)
results = self._net.detect(img)
#self.add_bounding_boxes(cv_img, results)
# bb_file = open('images/bounding_boxes{}'.format(msg.timestamp.coordinates[1]), 'wb')
# pickle.dump(results, bb_file)
# bb_file.close()
# output_msg = Message((msg.data, results), msg.timestamp)
runtime = (time.time() - start_time) * 1000
self._logger.info('Object detector {} runtime {}'.format(
self.name, runtime))
output_msg = Message((results, runtime), msg.timestamp)
self.get_output_stream(self._output_stream_name).send(output_msg)
def find_bounds_in_image(image_frame, darknet_model, class_label):
darknet_image_frame = pydarknet.Image(image_frame)
results = darknet_model.detect(
darknet_image_frame,
thresh=DETECTION_THRESHOLD,
hier_thresh=.5,
nms=.45) # todo: change this thresh-params thresh=0.01 #0.00051,
results_bounds = []
for category, score, bounds in results:
category = str(category.decode("utf-8"))
if category.lower() in class_label:
x, y, w, h = bounds
y1, y2 = int(y - h / 2), int(y + h / 2)
x1, x2 = int(x - w / 2), int(x + w / 2)
image_box_size_ratio = get_box_size_ratio(
y1, y2, x1, x2, image_shape=image_frame.shape)
if image_box_size_ratio < BOUND_RATIO_THRESHOLD:
results_bounds.append((y1, y2, x1, x2))
return results_bounds
def find_in_img(image_frame,
darknet_model,
class_labels,
thresh=0.5,
bound_size_thresh=0.5):
darknet_image_frame = pydarknet.Image(image_frame)
results = darknet_model.detect(
darknet_image_frame,
thresh=thresh, # 0.00051,
hier_thresh=.5,
nms=.45) # todo: change this thresh-params thresh=0.01 #0.00051,
results_bounds = []
for category, score, bounds in results:
category = str(category.decode("utf-8"))
if category.lower() in [label.lower() for label in class_labels]:
x, y, w, h = bounds
y1, y2 = int(y - h / 2), int(y + h / 2)
x1, x2 = int(x - w / 2), int(x + w / 2)
if box_size(y1, y2, x1, x2,
image_shape=image_frame.shape) < bound_size_thresh:
results_bounds.append((y1, y2, x1, x2))
return results_bounds
net = pydarknet.Detector(bytes(f"{DARKNET_LOCATION}/cfg/horsey-yolov3-tiny2.cfg", encoding="utf-8"),
bytes(f"{DARKNET_LOCATION}/backup/horsey-yolov3-tiny2_5500.weights", encoding="utf-8"),
0,
bytes(f"{DARKNET_LOCATION}/data/horsey-obj.data", encoding="utf-8"))
"""
# label a single image
if sys.argv[1] == "-i":
filepath = sys.argv[2]
img = cv2.imread(filepath)
# time the classification
start_time = time.time()
# Use YOLO Darknet to classify the image and store the results
img_darknet = pydarknet.Image(img)
results = net.detect(img_darknet)
end_time = time.time()
print(f"Time to classify frame: {end_time-start_time}")
for category, score, bounds in results:
# Bounds are BoundingBox coordinates, with x,y the centre, and w,h as width and height
x, y, w, h = bounds
# Label as found be YOLO CNN
label = str(category.decode("utf-8"))
# Draw Bounding Boxes around found RoIs, with different colours depending on found label/class
if label == "horse":
cv2.rectangle(img, (int(x - w / 2), int(y - h / 2)), (int(x + w / 2), int(y + h / 2)), (255, 0, 0),