def detect(self, image_cv) -> DetectorResponse:
#TO BE MODIFIED ----------------------
gender = False
#TO BE MODIFIED ----------------------
faces_bbox, faces_conf = cv.detect_face(image_cv)
bbox = []
label = []
conf = []
for bbox_item, conf_item in zip(faces_bbox, faces_conf):
startX = bbox_item[0]
startY = bbox_item[1]
endX = bbox_item[2]
endY = bbox_item[3]
c = float(conf_item)
l = 'person'
b = [int(startX), int(startY), int(endX), int(endY)]
bbox.append(b)
label.append(l)
conf.append(c)
# for faces_item, conf_item in zip(faces, conf):
# logger.debug("type={}, confidence={:.2f}%".format(faces_item,conf_item))
# (startX, startY) = faces_item[0], faces_item[1]
# (endX, endY) = faces_item[2], faces_item[3]
# c = float(conf_item)
# l = 'face'
# b = [int(startX), int(startY), int(endX), int(endY)]
# bbox.append(b)
# label.append(l)
# conf.append(c)
# if gender == True:
# face_crop = np.copy(image_cv[startY:endY, startX:endX])
# (gender_label_arr, gender_confidence_arr) = cv.detect_gender(face_crop)
# idx = np.argmax(gender_confidence_arr)
# gender_label = gender_label_arr[idx]
# gender_confidence = "{:.2f}%".format(gender_confidence_arr[idx] * 100)
# #obj['gender'] = gender_label
# #obj['gender_confidence'] = gender_confidence
model_response = DetectorResponse(self.get_model_name())
for l, c, b in zip(label, conf, bbox):
model_response.add(b, l, c, self.get_model_name())
return model_response
def detect(
self,
image_cv
) -> DetectorResponse:
bbox, label, conf = cv.detect_common_objects(image_cv)
model_response = DetectorResponse(self.get_model_name())
for l, c, b in zip(label, conf, bbox):
model_response.add(b,l,c,self.get_model_name())
return model_response
def detect(self, image_cv) -> DetectorResponse:
logger.debug("[REKOGNITION] request via boto3...")
imgHeight, imgWidth = image_cv.shape[:2]
pil_img = Image.fromarray(
image_cv
) # convert opencv frame (with type()==numpy) into PIL Image
stream = io.BytesIO()
pil_img.save(stream, format='JPEG') # convert PIL Image to Bytes
bin_img = stream.getvalue()
logger.debug("request via boto3...")
response = self.client.detect_labels(Image={'Bytes': bin_img})
#logger.debug("Reading {}".format(response))
logger.debug("...response received")
bbox = []
label = []
conf = []
for reko_label in response['Labels']:
for instance in reko_label['Instances']:
if reko_label['Name'] in self.rekognition_labels:
#extracting bounding box coordinates
left = int(imgWidth * instance['BoundingBox']['Left'])
top = int(imgHeight * instance['BoundingBox']['Top'])
width = int(imgWidth * instance['BoundingBox']['Width'])
height = int(imgHeight * instance['BoundingBox']['Height'])
x1 = left
y1 = top
x2 = left + width
y2 = top + height
l = self.__convert_label(reko_label['Name'])
c = float(reko_label['Confidence'] / 100)
b = [x1, y1, x2, y2]
bbox.append(b)
label.append(l)
conf.append(c)
model_response = DetectorResponse(self.get_model_name())
for l, c, b in zip(label, conf, bbox):
model_response.add(b, l, c, self.get_model_name())
return model_response
def detect(self, image_cv) -> DetectorResponse:
pil_image = Image.fromarray(
image_cv
) # convert opencv frame (with type()==numpy) into PIL Image
# Run inference.
ans = self.engine.detect_with_image(pil_image,
threshold=0.05,
keep_aspect_ratio=True,
relative_coord=False,
top_k=10)
bbox = []
conf = []
label = []
# Display result.
if ans:
for obj in ans:
#sample output
#score = 0.97265625
#box = [417.078184068203, 436.7141185646848, 2443.3632068037987, 1612.3385782686541]
b = obj.bounding_box.flatten().tolist()
b = [int(i) for i in b] #convert to int
c = float(obj.score)
l = 'person'
bbox.append(b)
conf.append(c)
label.append(l)
else:
logger.debug('No face detected!')
model_response = DetectorResponse(self.get_model_name())
for l, c, b in zip(label, conf, bbox):
model_response.add(b, l, c, self.get_model_name())
return model_response
def nms(self, merged: DetectorResponse, conf_threshold: float,
nms_threshold: float) -> (DetectorResponse, object):
start_total = datetime.datetime.now()
b, l, c, m = merged.get_blcm_vectors()
indices = cv2.dnn.NMSBoxes(b, c, conf_threshold, nms_threshold)
nms_out = DetectorResponse("nms")
for i in indices:
i = i[0]
nms_out.add(b[i], l[i], c[i], m[i])
for record in nms_out.data:
print("nms: {},{},{},{}".format(record.label, record.conf,
record.model_name, record.bbox))
stop_total = datetime.datetime.now()
total_time = stop_total - start_total
logger.info('TOTAL detection took {}'.format(total_time))
return nms_out, total_time
def merge(self,
batch_response: RunBatchResponse) -> (DetectorResponse, object):
logger.debug('merging...')
start_total = datetime.datetime.now()
merged = DetectorResponse("merged")
for response_item in batch_response.response_list:
merged.data += response_item.model_response.data
for record in merged.data:
print("merged: {},{},{},{}".format(record.label, record.conf,
record.model_name, record.bbox))
stop_total = datetime.datetime.now()
total_time = stop_total - start_total
logger.info('TOTAL detection took {}'.format(total_time))
return merged, total_time
def detect(self, image_cv) -> DetectorResponse:
pil_image = Image.fromarray(
image_cv
) # convert opencv frame (with type()==numpy) into PIL Image
# Run inference.
ans = self.engine.detect_with_image(pil_image,
threshold=0.05,
keep_aspect_ratio=True,
relative_coord=False,
top_k=10)
pre_bbox = []
pre_label = []
pre_conf = []
# Display result.
if ans:
for obj in ans:
if self.labels:
#sample output
#score = 0.97265625
#box = [417.078184068203, 436.7141185646848, 2443.3632068037987, 1612.3385782686541]
box = obj.bounding_box.flatten().tolist()
box = [int(i) for i in box] #convert to int
l = self.labels[obj.label_id]
c = float(obj.score)
b = box
# x = int(box[0])
# y = int(box[1])
# w = int(box[2] - box[0])
# h = int(box[3] - box[1])
# b = [x,y,w,h]
pre_bbox.append(b)
pre_label.append(l)
pre_conf.append(c)
# if c > conf_min:
# post_bbox.append(b)
# post_label.append(l)
# post_conf.append(c)
# else:
# logger.debug("DISCARDED as conf={} and threashold={}".format(c, conf_min))
# Perform non maximum suppression to eliminate redundant overlapping boxes with
# lower confidences.
#la prima soglia rappresenta la confidenza minima che accetti per un bbox
#ti pare di aver capito che la seconda soglia rappresenta
#l'area in % di sovrapposizione tra due bbox
indices = cv2.dnn.NMSBoxes(pre_bbox, pre_conf, 0.2, 0.3)
logger.debug("NMS indices {}".format(indices))
bbox = []
label = []
conf = []
for i in indices:
j = i[0]
bbox.append(pre_bbox[j])
label.append(pre_label[j])
conf.append(pre_conf[j])
else:
logger.debug('No object detected!')
model_response = DetectorResponse(self.get_model_name())
for l, c, b in zip(label, conf, bbox):
model_response.add(b, l, c, self.get_model_name())
return model_response
def detect(self, image_cv) -> DetectorResponse:
Height, Width = image_cv.shape[:2]
scale = 0.00392
blob = cv2.dnn.blobFromImage(image_cv,
scale, (416, 416), (0, 0, 0),
True,
crop=False)
self.net.setInput(blob)
outs = self.net.forward(self.__get_output_layers())
class_ids = []
confidences = []
boxes = []
conf_threshold = 0.5
nms_threshold = 0.4
for out in outs:
for detection in out:
scores = detection[5:]
class_id = np.argmax(scores)
confidence = scores[class_id]
if confidence > 0.5:
center_x = int(detection[0] * Width)
center_y = int(detection[1] * Height)
w = int(detection[2] * Width)
h = int(detection[3] * Height)
x = center_x - w / 2
y = center_y - h / 2
class_ids.append(class_id)
confidences.append(float(confidence))
boxes.append([x, y, w, h])
indices = cv2.dnn.NMSBoxes(boxes, confidences, conf_threshold,
nms_threshold)
bbox = []
label = []
conf = []
for i in indices:
i = i[0]
box = boxes[i]
x = box[0]
y = box[1]
w = box[2]
h = box[3]
bbox.append([
int(round(x)),
int(round(y)),
int(round(x + w)),
int(round(y + h))
])
label.append(str(self.classes[class_ids[i]]))
conf.append(float(confidences[i]))
model_response = DetectorResponse(self.get_model_name())
for l, c, b in zip(label, conf, bbox):
model_response.add(b, l, c, self.get_model_name())
return model_response