def infer_on_video(args):
plugin = Network()
plugin.load_model(args.m,args.d,CPU_EXTENSION)
b,c,h,w = plugin.get_input_shape()
# print("Required shape:",b,c,h,w)
if args.i == 'CAM':
args.i = 0
cap = cv2.VideoCapture(args.i)
cap.open(args.i)
while cap.isOpened():
flag,frame = cap.read()
if not flag:
break
frame = cv2.resize(frame,(w,h))
r_frame = preprocessing(frame,c,h,w)
result = get_result(r_frame,plugin)
output = bounding_boxes(frame,result)
cv2.imshow("input", output)
if cv2.waitKey(1000 // int(cap.get(5))) == 27:
break
cap.release()
cv2.destroyAllWindows()
def __init__(self, model, device, cpu_ext, mqtt_server):
#required arguments
self.model = model
self.device = device
self.cpu_ext = cpu_ext
self.mqtt_client = mqtt_server
#argument for other functions
self.prob_threshold = None
self.prob_iou = None
self.input_ = None
self._out_name = None
self.out_serveronly = False
self.classes = None
self.frames = {}
self.prev_boxes = {}
self.curr_boxes = {}
# self.cuur_out = []
self.tolerate = {}
self.total_detect = 0
self.avg_time = 0
self.async_ = None
# Initialise the class
self.infer_network = Network()
### TODO: Load the model through `infer_network` ###
self.infer_network.load_model(self.model, self.device, self.cpu_ext)
#check if server is available
self.check_server()
def __init__(self,
model_name,
device='CPU',
extensions=None,
threshold=0.60):
self.threshold = threshold
self.network = Network(model_name, device, extensions)
def infer_on_stream(args, client):
"""
Initialize the inference network, stream video to network,
and output stats and video.
:param args: Command line arguments parsed by `build_argparser()`
:param client: MQTT client
:return: None
"""
# Initialise the class
infer_network = Network()
# Set Probability threshold for detections
## prob_threshold = args.prob_threshold
single_image_mode = False
cur_request_id = 0
last_count = 0
total_count = 0
start_time = 0
### TODO: Load the model through `infer_network` ###
infer_network = Network()
n, c, h, w = infer_network.load_model(args.model, args.device, 1, 1,
cur_request_id, args.cpu_extension)[1]
def infer_on_stream(args, client):
"""
Initialize the inference network, stream video to network,
and output stats and video.
:param args: Command line arguments parsed by `build_argparser()`
:param client: MQTT client
:return: None
"""
# Initialise the class
infer_network = Network()
# Set Probability threshold for detections
prob_threshold = args.prob_threshold
### TODO: Load the model through `infer_network` ###
inference_network.load_model(args.model, args.device, CPU_EXTENSION, num_requests=0)
_, _, inh, inw = infer_network.get_input_shape()
### TODO: Handle the input stream ###
try:
cap = cv2.VideoCapture(args.input)
except FileNotFoundError:
print("Cannot locate video file: "+ args.input)
except Exception as e:
print("Something else went wrong with the video file: ", e)
def __init__(self, model, device="CPU", extensions=None):
"""
set instance variables
"""
self.model_xml = model
self.device = device
self.extensions = extensions
self.infer_network = Network()
def infer_on_stream(args, client):
"""
Initialize the inference network, stream video to network,
and output stats and video.
:param args: Command line arguments parsed by `build_argparser()`
:param client: MQTT client
:return: None
"""
# Initialise the class
infer_network = Network()
# Set Probability threshold for detections
prob_threshold = args.prob_threshold
### TODO: Load the model through `infer_network` ###
plugin = Network()
plugin.load_model(model, args.d, CPU_EXTENSION)
net_input_shape = plugin.get_input_shape()
### TODO: Handle the input stream ###
cap = cv2.VideoCapture(args.i)
cap.open(args.i)
width = int(cap.get(3))
height = int(cap.get(4))
### TODO: Loop until stream is over ###
while cap.isOpened():
### TODO: Read from the video capture ###
flag, frame = cap.read()
if not flag:
break
key_pressed = cv2.waitKey(60)
### TODO: Pre-process the image as needed ###
p_frame = cv2.resize(frame, (net_input_shape[3], net_input_shape[2]))
p_frame = p_frame.transpose((2,0,1))
p_frame = p_frame.reshape(1, *p_frame.shape)
### TODO: Start asynchronous inference for specified request ###
plugin.async_inference(p_frame)
### TODO: Wait for the result ###
if plugin.wait() == 0:
### TODO: Get the results of the inference request ###
result = plugin.extract_output()
### TODO: Extract any desired stats from the results ###
print("result :",result)
### TODO: Calculate and send relevant information on ###
### current_count, total_count and duration to the MQTT server ###
### Topic "person": keys of "count" and "total" ###
client.publish("person", json.dumps({"count": 0,"total":0}))
### Topic "person/duration": key of "duration" ###
client.publish("person/duration", json.dumps({"duration": 0}))
### TODO: Send the frame to the FFMPEG server ###
sys.stdout.buffer.write(out_frame)
sys.stdout.flush()
def load_models(video_caps, args):
"""
Load the required models
:param video_caps: List of VideoCap objects
:param args: Command line arguments
:return: None
"""
global check_feed_type
# Check if one of the feed type is "shopper". If yes, load the face, head pose and mood detection model
if check_feed_type[0]:
infer_network_face = Network()
infer_network_pose = Network()
infer_network_mood = Network()
tag_face = {"VPU_HDDL_GRAPH_TAG": "tagFace"}
tag_pose = {"VPU_HDDL_GRAPH_TAG": "tagPose"}
tag_mood = {"VPU_HDDL_GRAPH_TAG": "tagMood"}
nchw_fd = load_model_device(infer_network_face, args.facemodel,
args.facedevice, 1, 1, 2,
args.cpu_extension, tag_face)
nchw_hp = load_model_device(infer_network_pose, args.posemodel,
args.posedevice, 1, 3, 2,
args.cpu_extension, tag_pose)
nchw_md = load_model_device(infer_network_mood, args.moodmodel,
args.mooddevice, 1, 1, 2,
args.cpu_extension, tag_mood)
if check_feed_type[2]:
infer_network = Network()
tag_obj = {"VPU_HDDL_GRAPH_TAG": "tagMobile"}
nchw = load_model_device(infer_network, args.objmodel,
args.objectdevice, 1, 1, 2,
args.cpu_extension, tag_obj)
if check_feed_type[1]:
infer_network_person = Network()
tag_person = {"VPU_HDDL_GRAPH_TAG": "tagPerson"}
nchw_pr = load_model_device(infer_network_person, args.personmodel,
args.persondevice, 2, 1, 2,
args.cpu_extension, tag_person)
for video_cap in video_caps:
if video_cap.type == 'shopper':
video_cap.infer_network = infer_network_face
video_cap.infer_network_hp = infer_network_pose
video_cap.infer_network_md = infer_network_mood
video_cap.nchw.extend(nchw_fd)
video_cap.nchw_hp.extend(nchw_hp)
video_cap.nchw_md.extend(nchw_md)
if video_cap.type == 'shelf':
video_cap.infer_network = infer_network
video_cap.nchw.extend(nchw)
if video_cap.type == 'traffic':
video_cap.infer_network = infer_network_person
video_cap.nchw.extend(nchw_pr)
def load_model(self):
'''
TODO: You will need to complete this method.
This method is for loading the model to the device specified by the user.
If your model requires any Plugins, this is where you can load them.
'''
self.infer_network = Network()
self.infer_network.load_model(self.model_xml, self.device,
self.extensions)
def __init__(self, model_name, device='CPU', extensions=None):
'''
TODO: Use this to set your instance variables.
'''
self.model_xml = model_name
self.device = device
self.extensions = extensions
# Initialise the class
self.infer_network = Network()
def infer_on_video(args):
plugin = Network()
plugin.load_model(args.m,args.d,None)
net_input_shape = plugin.get_input_shape()
cap = cv2.VideoCapture(args.i)
cap.open(args.i)
counter = 0
incidence_flag = False
while cap.isOpened():
flag,frame = cap.read()
if not flag:
break
counter += 1
r_frame = cv2.resize(frame, (net_input_shape[3],net_input_shape[2]))
r_frame = r_frame.transpose((2,0,1))
r_frame = r_frame.reshape(1,*r_frame.shape)
plugin.async_inference(r_frame)
if plugin.wait() == 0:
result = plugin.extract_output()
incidence_flag = incidence(result,counter,incidence_flag)
cap.release()
cv2.destroyAllWindows()
def infer_on_stream(cap, out, width, height, args):
infer_network = Network()
infer_network.load_model(args.model, args.device, args.cpu_extension)
net_input_shape = infer_network.get_input_shape()
while cap.isOpened():
flag, frame = cap.read()
if not flag:
break
p_frame = cv2.resize(frame, (net_input_shape[3], net_input_shape[2]))
p_frame = p_frame.transpose((2, 0, 1))
p_frame = p_frame.reshape(1, *p_frame.shape)
infer_network.exec_net(p_frame)
if infer_network.wait() == 0:
result = infer_network.get_output()
frame, current_count = draw_boxes(frame, result, width, height,
float(args.prob_threshold))
send_update = update_count(current_count,
cap.get(cv2.CAP_PROP_POS_MSEC) / 1000)
if send_update:
print(current_count, total_count, duration)
out.write(frame)
def infer_on_video(args):
### Initialize the Inference Engine
plugin = Network()
### Load the network model into the IE
plugin.load_model(args.m, args.d)
input_shape = plugin.get_input_shape()
# Get and open video capture
cap = cv2.VideoCapture(args.i)
cap.open(args.i)
# Grab the shape of the input
width = int(cap.get(3))
height = int(cap.get(4))
# Create a video writer for the output video
# The second argument should be `cv2.VideoWriter_fourcc('M','J','P','G')`
# on Mac, `0x00000021` on Udacity IDE, and `0x00000021` on Linux
out = cv2.VideoWriter('out.mp4', 'cv2.VideoWriter_fourcc("M","J","P","G")',
30, (width, height))
# Process frames until the video ends, or process is exited
while cap.isOpened():
# Read the next frame
flag, frame = cap.read()
if not flag:
break
key_pressed = cv2.waitKey(60)
### Pre-process the frame
image = cv2.resize(frame, (input_shape[3], input_shape[2]))
image = image.transpose((2, 0, 1))
image = image.reshape(1, 3, input_shape[2], input_shape[3])
### Perform inference on the frame
plugin.async_inference(image)
### Get the output of inference
if plugin.wait() == 0:
result = plugin.extract_output()
### Update the frame to include detected bounding boxes
for box in result[0][0]:
if box[2] >= float(args.conf):
x_min = int(box[3] * width)
y_min = int(box[4] * height)
x_max = int(box[5] * width)
y_max = int(box[6] * height)
cv2.rectangle(frame, (x_min, y_min), (x_max, y_max),
get_color(args.color), 1)
# Write out the frame
out.write(frame)
# Break if escape key pressed
if key_pressed == 27:
break
# Release the out writer, capture, and destroy any OpenCV windows
out.release()
cap.release()
cv2.destroyAllWindows()
def infer_on_stream():
# Initialise the class
infer_network = Network()
infer_network.load_model(
"./models/mobilenet_ssd_pedestrian_detection/MobileNetSSD_deploy10695.xml",
"CPU", CPU_EXTENSION)
net_input_shape = infer_network.get_input_shape()
### TODO: Handle the input stream ###
img = cv2.imread('./frame1.jpg', 0)
img = cv2.resize(img, (net_input_shape[3], net_input_shape[2]))
imgProcessed = img - 127.5
imgProcessed = imgProcessed * 0.007843
imgProcessed = imgProcessed.astype(np.float32)
infer_network.exec_net(imgProcessed)
if infer_network.wait() == 0:
### TODO: Get the results of the inference request ###
width, height = imgProcessed.shape[:2]
result = infer_network.get_output()
h = img.shape[0]
w = img.shape[1]
box = result[0, 0, :, 3:7] * np.array([w, h, w, h])
cls = result[0, 0, :, 1]
conf = result[0, 0, :, 2]
for i in range(len(box)):
aR = abs(box[i][2] - box[i][0]) * (box[i][3] - box[i][1])
if conf[i] > 0.25 and aR < 30000:
cv2.rectangle(img, (int(box[i][0]), int(box[i][1])),
(int(box[i][2]), int(box[i][3])), (0, 255, 0))
cv2.imwrite("frameProcessed.jpg", img)
def infer_on_camera(args):
# Convert the args for color and confidence
args.c = convert_color(args.c)
args.ct = float(args.ct)
# Initialize the Inference Engine
plugin = Network()
# Load the network model into the IE
plugin.load_model(args.m, args.d, CPU_EXTENSION)
# Get input shape
net_input_shape = plugin.get_input_shape()
# Get and open video capture
cap = cv2.VideoCapture(0)
cap.open(0) # 0 for default camera
# Grab the shape of the input
width = int(cap.get(3))
height = int(cap.get(4))
# Process frames until video end or process is exited
while cap.isOpened():
flag, frame = cap.read()
if not flag:
break
key_pressed = cv2.waitKey(60)
# Preprocess the frame
p_frame = preprocessing(frame, 384, 672)
# Perform inference on the frame
plugin.async_inference(p_frame)
# Get the output of the inference
if plugin.wait() == 0:
result = plugin.extract_output()
# Update the frame to include detected bounding boxes
frame = draw_boxes(frame, result, args, width, height)
cv2.imshow("frame", frame)
if key_pressed == 27:
break
def infer_on_image(args):
print('INFER ON IMAGE')
# Convert the args for confidence
args.ct = float(args.ct)
### Initialize the Inference Engine
plugin = Network()
### Load the network model into the IE
plugin.load_model(args.m, args.d)
net_input_shape = plugin.get_input_shape()
# Read the input image
image = cv2.imread(args.i)
h, w = net_input_shape[2], net_input_shape[3]
### Preprocess the input image
preprocessed_image = preprocessing(image, h, w)
### Perform inference on the frame
plugin.async_inference(preprocessed_image)
### Get the output of inference
if plugin.wait() == 0:
output = plugin.extract_output()
image = draw_boxes(image, output, args, w, h)
cv2.imwrite(args.o, image)
def pred_at_edge(input_img):
# Initialize the Inference Engine
plugin = Network()
# Load the network model into the IE
plugin.load_model(MODEL, "CPU", CPU_EXTENSION)
net_input_shape = plugin.get_input_shape()
# Reading input image
img = cv2.imread(input_img, cv2.IMREAD_COLOR)
# Pre-process the image
expand_img = preprocessing(img, net_input_shape[2], net_input_shape[3])
final_img = np.expand_dims(expand_img, axis=0)
# Perform inference on the image
plugin.async_inference(final_img)
# Get the output of inference
if plugin.wait() == 0:
results = plugin.extract_output()
pred = np.argmax(results)
disease = SKIN_CLASSES[pred]
accuracy = results[0][pred]
print(disease, accuracy)
return disease, accuracy
def infer_on_video(args):
# Convert the args for color and confidence
args.c = convert_color(args.c)
args.ct = float(args.ct)
### TODO: Initialize the Inference Engine
plugin = Network()
### TODO: Load the network model into the IE
plugin.load_model(args.m, args.d, CPU_EXTENSION)
net_input_shape = plugin.get_input_shape()
# Get and open video capture
cap = cv2.VideoCapture(args.i)
cap.open(args.i)
# Grab the shape of the input
width = int(cap.get(3))
height = int(cap.get(4))
# Create a video writer for the output video
# The second argument should be `cv2.VideoWriter_fourcc('M','J','P','G')`
# on Mac, and `0x00000021` on Linux
out = cv2.VideoWriter(FILE_OUTPUT,
cv2.VideoWriter_fourcc('m', 'p', '4', 'v'), 30,
(width, height))
# Process frames until the video ends, or process is exited
while cap.isOpened():
# Read the next frame
flag, frame = cap.read()
if not flag:
break
key_pressed = cv2.waitKey(60)
### TODO: Pre-process the frame
p_frame = cv2.resize(frame, (net_input_shape[3], net_input_shape[2]))
p_frame = p_frame.transpose((2, 0, 1))
p_frame = p_frame.reshape(1, *p_frame.shape)
### TODO: Perform inference on the frame
plugin.async_inference(p_frame)
### TODO: Get the output of inference
if plugin.wait() == 0:
result = plugin.extract_output()
### TODO: Update the frame to include detected bounding boxes
frame = draw_boxes(frame, result, args, width, height)
# Write out the frame
out.write(frame)
# Break if escape key pressed
if key_pressed == 27:
break
# Release the out writer, capture, and destroy any OpenCV windows
out.release()
cap.release()
cv2.destroyAllWindows()
def infer_on_video(args):
### TODO: Initialize the Inference Engine
net = Network()
### TODO: Load the network model into the IE
net.load_model(model=args.m, device=args.d, cpu_extension=CPU_EXTENSION)
net_input_shape = net.get_input_shape()
# Get and open video capture
cap = cv2.VideoCapture(args.i)
cap.open(args.i)
# Grab the shape of the input
width = int(cap.get(3))
height = int(cap.get(4))
# Create a video writer for the output video
# The second argument should be `cv2.VideoWriter_fourcc('M','J','P','G')`
# on Mac, and `0x00000021` on Linux
fourcc = cv2.VideoWriter_fourcc(*"mp4v")
out = cv2.VideoWriter("out.mp4", fourcc, 30, (width, height))
# Process frames until the video ends, or process is exited
while cap.isOpened():
# Read the next frame
flag, frame = cap.read()
if not flag:
break
key_pressed = cv2.waitKey(60)
### TODO: Pre-process the frame
# Taken from lesson 2: preprocess_input.py
p_frame = cv2.resize(frame, (net_input_shape[3], net_input_shape[2]))
p_frame = p_frame.transpose((2, 0, 1))
p_frame = p_frame.reshape(1, *p_frame.shape)
### TODO: Perform inference on the frame
net.async_inference(image=p_frame)
### TODO: Get the output of inference
if net.wait() == 0:
result = net.extract_output()
### TODO: Update the frame to include detected bounding boxes
try:
frame = draw_boxes(frame, result, args, width, height)
except Exception as e:
print(str(e))
# Write out the frame
else:
out.write(frame)
# Break if escape key pressed
if key_pressed == 27:
break
# Release the out writer, capture, and destroy any OpenCV windows
out.release()
cap.release()
cv2.destroyAllWindows()
def main():
args = build_argparser().parse_args()
scores = list()
scores_wout_mispredictions = list()
inference_time = list()
network = Network()
network.load_model(args.model,
cpu_extension=args.cpu_extension,
device=args.device)
net_input_shape = network.get_input_shape()
if args.image_directory is not None:
if os.path.exists(args.image_directory):
for root, dirs, files in os.walk(args.image_directory,
topdown=False):
counter = 0
for name in files:
if counter >= args.maximum_images:
break
frame = cv2.imread(os.path.join(root, name))
res, inf_time = post_conversion_benchmark(
frame, args.model, args.cpu_extension, args.device,
args.prob_threshold, args.iou_threshold, network,
net_input_shape)
if res > 0:
scores_wout_mispredictions.append(res)
scores.append(res)
inference_time.append(inf_time)
counter += 1
else:
res, inf_time = post_conversion_benchmark(
args.input_img, args.model, args.cpu_extension, args.device,
args.prob_threshold, args.iou_threshold, network, net_input_shape)
if res > 0:
scores_wout_mispredictions.append(res)
scores.append(res)
inference_time.append(inf_time)
print("Average score across all images: " + str(np.mean(scores)))
print("Max score across all images: " + str(np.max(scores)))
print("Average inference time: " + str(np.mean(inference_time)) + "ms")
print("Average score disregarding mis-predictions: " +
str(np.mean(scores_wout_mispredictions)))
print("Minimum score disregarding mis-predictions: " +
str(np.min(scores_wout_mispredictions)))
def load_models(video_caps):
"""
Load the required models
:param video_caps: List of VideoCap objects
:return: None
"""
global check_feed_type
plugin = None
face_device = os.environ['FACE_DEVICE'] if 'FACE_DEVICE' in os.environ.keys() else "CPU"
mood_device = os.environ['MOOD_DEVICE'] if 'MOOD_DEVICE' in os.environ.keys() else "CPU"
pose_device = os.environ['POSE_DEVICE'] if 'POSE_DEVICE' in os.environ.keys() else "CPU"
obj_device = os.environ['OBJ_DEVICE'] if 'OBJ_DEVICE' in os.environ.keys() else "CPU"
cpu_extension = os.environ['CPU_EXTENSION'] if 'CPU_EXTENSION' in os.environ.keys() else None
face_model = os.environ['FACE_MODEL'] if 'FACE_MODEL' in os.environ.keys() else None
pose_model = os.environ['POSE_MODEL'] if 'POSE_MODEL' in os.environ.keys() else None
mood_model = os.environ['MOOD_MODEL'] if 'MOOD_MODEL' in os.environ.keys() else None
obj_model = os.environ['OBJ_MODEL'] if 'OBJ_MODEL' in os.environ.keys() else None
# Check if one the feed type is "shopper". If yes, load the face, head pose and mood detection model
if check_feed_type[0]:
assert face_model, 'Please specify the path to face detection model using the environment variable FACE_MODEL'
assert pose_model, 'Please specify the path to head pose model using the environment variable POSE_MODEL'
assert mood_model, 'Please specify the path to mood detection model using the environment variable MOOD_MODEL'
infer_network_face = Network()
infer_network_pose = Network()
infer_network_mood = Network()
nchw_fd = load_model_device(infer_network_face, face_model, face_device, 1, 1, 0, cpu_extension)
nchw_hp = load_model_device(infer_network_pose, pose_model, pose_device, 1, 3, 0, cpu_extension)
nchw_md = load_model_device(infer_network_mood, mood_model, mood_device, 1, 1, 0, cpu_extension)
# Check if one the feed type is "traffic" or "shelf". If yes, load the mobilenet-ssd model
if check_feed_type[1] or check_feed_type[2]:
assert obj_model, 'Please specify the path to object detection model using the environment variable OBJMODEL'
infer_network = Network()
nchw = load_model_device(infer_network, obj_model, obj_device, 1, 1, 2, cpu_extension)
for video_cap in video_caps:
if video_cap.type == 'shopper':
video_cap.infer_network = infer_network_face
video_cap.infer_network_hp = infer_network_pose
video_cap.infer_network_md = infer_network_mood
video_cap.nchw.extend(nchw_fd)
video_cap.nchw_hp.extend(nchw_hp)
video_cap.nchw_md.extend(nchw_md)
if video_cap.type == 'traffic' or video_cap.type == 'shelf':
video_cap.infer_network = infer_network
video_cap.nchw.extend(nchw)
def infer_on_video(args):
### TODO: Initialize the Inference Engine
plugin = Network()
### TODO: Load the network model into the IE
plugin.load_model(args.m, args.d, CPU_EXTENSION)
net_input_shape = plugin.get_input_shape()
print('input shape of net format fct of model use \n', net_input_shape)
#[B,C,H,W]
# Get and open video capture
cap = cv2.VideoCapture(args.i)
cap.open(args.i)
print("capture 3: \n ", cap.get(3)) # video width
print("capture 4: \n ", cap.get(4)) # video heigth
# Grab the shape of the input
width = int(cap.get(3))
height = int(cap.get(4))
# Create a video writer for the output video
# The second argument should be `cv2.VideoWriter_fourcc('M','J','P','G')`
# on Mac, and `0x00000021` on Linux
out = cv2.VideoWriter('out.mp4', 0x00000021, 30, (width, height))
# Process frames until the video ends, or process is exited
while cap.isOpened():
# Read the next frame
flag, frame = cap.read()
if not flag:
break
key_pressed = cv2.waitKey(60)
### TODO: Pre-process the frame(each frame is an image)
preproced_frame = cv2.resize(frame,
(net_input_shape[3], net_input_shape[2]))
preproced_frame = preproced_frame.transpose((2, 0, 1))
preproced_frame = preproced_frame.reshape(1, *preproced_frame.shape)
### TODO: Perform inference on the frame
plugin.async_inference(preproced_frame)
### TODO: Get the output of inference
if plugin.wait() == 0:
result = plugin.extract_output()
#print('results: \n',result)
### TODO: Update the frame to include detected bounding boxes
frame = draw_boxes(frame, result, args, width, height)
# Write out the frame
out.write(frame)
# Break if escape key pressed
if key_pressed == 27:
break
# Release the out writer, capture, and destroy any OpenCV windows
out.release()
cap.release()
cv2.destroyAllWindows()
def emotion_detection(emotions_model, frame, result, args, width, height):
"""
Detect the emotion of the faces of a frame.
"""
# Initialize the Inference Engine
plugin_emotions_detection = Network()
# Load the network models into the IE
plugin_emotions_detection.load_model(emotions_model, args.d, CPU_EXTENSION)
net_input_shape_ed = plugin_emotions_detection.get_input_shape()
for box in result[0][0]:
conf = box[2]
if conf >= args.ct:
# calculate the rectangle box margins
x_min = max(int(box[3] * width), 0)
y_min = max(int(box[4] * height), 0)
x_max = min(int(box[5] * width), width)
y_max = min(int(box[6] * height), height)
# crop the image for emotion detection
cropped_frame = frame[y_min:y_max, x_min:x_max]
if cropped_frame.shape[0] and cropped_frame.shape[1]:
# Draw rectangle box on the input
cv2.rectangle(frame, (x_min, y_min), (x_max, y_max), (0, 0, 255), 1)
print('cropped frame: ', cropped_frame.shape)
# Preprocess the cropped image
p_frame_ed = preprocessing(cropped_frame, net_input_shape_ed[2], net_input_shape_ed[3])
# Perform inference on the frame to detect emotion
plugin_emotions_detection.async_inference(p_frame_ed)
if plugin_emotions_detection.wait() == 0:
result_ed = plugin_emotions_detection.extract_output()
# Get the emotions class
emotion_class_id = np.argmax(result_ed)
emotion_class = EMOTIONS[emotion_class_id]
print('emotion detected:', emotion_class)
# # Crate a rectangle box to display emotion text
# sub_img = frame[y_min:y_min+20, x_min:x_max]
# white_rect = np.ones(sub_img.shape, dtype=np.uint8) * 255
# res = cv2.addWeighted(sub_img, 0.5, white_rect, 0.5, 1.0)
# # Putting the image back to its position
# frame[y_min:y_min+20, x_min:x_max] = res
# Create a rectangle to display the predicted emotion
cv2.rectangle(frame, (x_min, y_min), (x_max, y_min + 20), (51, 255, 196), cv2.FILLED)
cv2.putText(frame, emotion_class, (x_min + 5, y_min + 15), cv2.FONT_HERSHEY_SIMPLEX, 0.5,
(255, 0, 0), 2)
return frame
def infer_on_video(args, model):
### TODO: Connect to the MQTT server
# Initialize the Inference Engine
plugin = Network()
# Load the network model into the IE
plugin.load_model(model, args.d, CPU_EXTENSION)
net_input_shape = plugin.get_input_shape()
# Get and open video capture
cap = cv2.VideoCapture(args.i)
cap.open(args.i)
# Grab the shape of the input
width = int(cap.get(3))
height = int(cap.get(4))
# Process frames until the video ends, or process is exited
while cap.isOpened():
# Read the next frame
flag, frame = cap.read()
if not flag:
break
key_pressed = cv2.waitKey(60)
# Pre-process the frame
p_frame = cv2.resize(frame, (net_input_shape[3], net_input_shape[2]))
p_frame = p_frame.transpose((2, 0, 1))
p_frame = p_frame.reshape(1, *p_frame.shape)
# Perform inference on the frame
plugin.async_inference(p_frame)
# Get the output of inference
if plugin.wait() == 0:
result = plugin.extract_output()
# Draw the output mask onto the input
out_frame, classes = draw_masks(result, width, height)
class_names = get_class_names(classes)
speed = randint(50, 70)
### TODO: Send the class names and speed to the MQTT server
### Hint: The UI web server will check for a "class" and
### "speedometer" topic. Additionally, it expects "class_names"
### and "speed" as the json keys of the data, respectively.
### TODO: Send frame to the ffmpeg server
# Break if escape key pressed
if key_pressed == 27:
break
# Release the capture and destroy any OpenCV windows
cap.release()
cv2.destroyAllWindows()
def infer_on_video():
speak('Hello! I am Alicia. Please wait while I boot up the system.')
# Initialize the Inference Engine
plugin = Network()
class_names = []
# Load the network model into the IE
plugin.load_model(MODEL, DEVICE)
net_input_shape = plugin.get_input_shape()
camera = PiCamera()
camera.resolution = (640, 480)
camera.framerate = 32
rawCapture = PiRGBArray(camera, size=(640, 480))
speak('System booted.')
# allow the camera to warmup
time.sleep(0.1)
for frame in camera.capture_continuous(rawCapture,
format="bgr",
use_video_port=True):
# grab the raw image
image = frame.array
key_pressed = cv2.waitKey(60)
# Pre-process the frame
p_frame = cv2.resize(image, (net_input_shape[3], net_input_shape[2]))
p_frame = p_frame.transpose((2, 0, 1))
p_frame = p_frame.reshape(1, *p_frame.shape)
# Perform inference on the frame
plugin.async_inference(p_frame)
# Get the output of inference
if plugin.wait() == 0:
result = plugin.extract_output()
classes = np.transpose(result[0])[1]
# classes = np.unique(np.transpose(result[0])[1])
old_class_names = class_names
class_names = get_class_names(classes)
# publish the new environment only if something changes
if class_names and class_names != old_class_names:
speak_string = ''.join(class_names)
speak(speak_string)
print(class_names)
print("----------------------")
# clear the stream in preparation for the next frame
rawCapture.truncate(0)
# Break if escape key pressed
if key_pressed == 27:
break
def perform_inference(args):
'''
Performs inference on an input image, given a model.
'''
# Create a Network for using the Inference Engine
plugin = Network()
# Load the network model into the IE
plugin.load_model(args.m, args.d, args.c)
net_input_shape = plugin.get_input_shape()
# Read the input image
image = cv2.imread(args.i)
# Preprocess the input image
preprocessed_image = preprocessing(image, net_input_shape[3],
net_input_shape[4])
# Perform synchronous inference on the image
plugin.sync_inference(preprocessed_image)
# Obtain the output of the inference request
output = plugin.extract_output()
processed_output = handle_asl(output)
# Create an output image based on network
output_image = create_output_image(image, processed_output)
# Save down the resulting image
cv2.imwrite("outputs/output.png", output_image)
def perform_inference(args):
'''
Performs inference on an input image, given a model.
'''
# Create a Network for using the Inference Engine
plugin = Network()
# Load the network model into the IE
plugin.load_model(args.m, args.d, args.c)
net_input_shape = plugin.get_input_shape()
# Read the input image
image = Image.open(args.i)
# Preprocess the input image
preprocessed_image = preprocessing(image)
# Perform synchronous inference on the image
plugin.sync_inference(preprocessed_image)
# Obtain the output of the inference request
output = plugin.extract_output()
processed_output = handle_output(output)
processed_output = MODEL_OUTPUT[processed_output].split()[0]
ret_string = get_nutritional_data(processed_output, args.id, args.k)
# Create an output image based on network
output_image = create_output_image(ret_string)
# Save down the resulting image
cv2.imwrite("../outputs/output.png", output_image)
def perform_facerecognition(face,model): plugin=Network() plugin.load_model(model=model) b,c,h,w=plugin.get_input_shape() p_image=preprocessing(face,h,w) plugin.async_inference(p_image) status=plugin.wait() if status==0: result=plugin.extract_output() candidate_embedding=result[0] return candidate_embedding
def main():
args = build_argparser().parse_args()
# Connect to the MQTT server
client = connect_mqtt()
# Perform inference on the input stream
model = args.model
Device = args.Device
CPU_extension = args.cpu_extension
net = Network()
net.load_model(model, Device, CPU_extension)
net_input_shape = net.get_input_shape()['image_tensor']
net_shape = (net_input_shape[3], net_input_shape[2])
infer_on_stream(net, args, client, net_shape)
def infer(imarray):
model = "asl-recognition-0003.xml"
inet = Network.net(model)
exec_net = Network.load_model(inet, imarray, 'CPU')
input_blob = next(iter(exec_net.inputs))
input_layer = inet.inputs[input_blob].shape
##### asynch
asy_net = Network.async_inference(exec_net, imarray, input_blob)
output_blob = next(iter(exec_net.outputs))
enc_net = exec_net.requests[0].outputs[output_blob]
###### synch
#syn_net = Network.inf_(exec_net, imarray, input_blob)
#enc_net = Network.extract_output(syn_net, exec_net)
code = (np.argmax(enc_net))
return code
if algo == 'orig':
edges = GPT_learn_original (tree, data, variables)
elif algo == 'mod':
edges = GPT_learn_modified (tree, data, variables)
else:
print 'wrong algo'
sys.exit (-1)
if arg == "predict":
import networkx as nx
g = nx.DiGraph ()
g.add_edges_from (edges)
from inference import Network, get_factors_BN
n = Network (get_factors_BN (g, variables, data))
with open ('test_data.txt') as f:
f.readline () #exclude the first row
for l in f.readlines ():
values = l.strip ().split ()
assignment = dict(zip(variables, values))
print n.total_joint_prob (assignment)
elif arg == "graphviz":
viz_str = viz_tree_str (edges, variables, directed = True)
print viz_str
elif arg == "sorted_edges":
for v1, v2 in sort_edges (edges, list (permutations (variables, 2)), edge_weights):
print "%s %s" %(v1, v2)
else:
