def main(args):
prediction_client = PredictionClient(args.address, args.port)
if DEVICE_CONNECTION_STRING and not args.suppress_messages:
iothub_client = IoTHubClient(DEVICE_CONNECTION_STRING, PROTOCOL)
while True:
for image in os.listdir(args.image_dir):
# score image
try:
start_time = time.time()
results = prediction_client.score_file(
path=os.path.join(args.image_dir, image),
input_name=args.input_tensors,
outputs=args.output_tensors)
inference_time = (time.time() - start_time) * 1000
result = 'cat' if results[0] > results[1] else 'dog'
msg_string = "(%.3f ms) The image %s is a '%s'." % (
inference_time, image, result)
print(msg_string)
except:
tb = traceback.format_exc()
if "StatusCode.UNAVAILABLE" in tb:
msg_string = "Unable to inference because AzureML host container is not done flashing the FPGA. If still not available in 5 minutes, check logs of module."
elif "Object reference not set to an instance of an object" in tb:
msg_string = "Unable to inference because the names of the input and output tensors used for scoring are incorrect.\n" + \
"Please update the docker CMD arguments to include the correct --input-tensors and --output-tensors parameters."
else:
msg_string = "Unable to inference for unknown reason. See stack trace below:\n{}".format(
tb)
print(msg_string)
print("Trying again in {} seconds...".format(args.wait))
if DEVICE_CONNECTION_STRING and not args.suppress_messages:
send_iothub_message(iothub_client, msg_string)
time.sleep(args.wait)
def main(args):
prediction_client = PredictionClient(args.address, args.port)
if DEVICE_CONNECTION_STRING and not args.suppress_messages:
iothub_client = IoTHubClient(DEVICE_CONNECTION_STRING, PROTOCOL)
classes_entries = requests.get(
"https://raw.githubusercontent.com/Lasagne/Recipes/master/examples/resnet50/imagenet_classes.txt"
).text.splitlines()
while True:
for image in os.listdir(args.image_dir):
# score image
try:
start_time = time.time()
results = prediction_client.score_file(
path=os.path.join(args.image_dir, image),
input_name=args.input_tensors,
outputs=args.output_tensors)
inference_time = (time.time() - start_time) * 1000
# map results [class_id] => [confidence]
results = enumerate(results)
# sort results by confidence
sorted_results = sorted(results,
key=lambda x: x[1],
reverse=True)
top_result = sorted_results[0]
msg_string = "(%.0f ms) The image %s was classified as %s with confidence %s." % (
inference_time, os.path.join(args.image_dir, image),
classes_entries[top_result[0]], top_result[1])
print(msg_string)
except:
tb = traceback.format_exc()
if "StatusCode.UNAVAILABLE" in tb:
msg_string = "Unable to inference because AzureML host container is not done flashing the FPGA. If still not available in 5 minutes, check logs of module."
elif "Object reference not set to an instance of an object" in tb:
msg_string = "Unable to inference because the names of the input and output tensors used for scoring are incorrect.\n" + \
"Please update the docker CMD arguments to include the correct --input-tensors and --output-tensors parameters."
else:
msg_string = "Unable to inference for unknown reason. See stack trace below:\n{}".format(
tb)
print(msg_string)
print("Trying again in {} seconds...".format(args.wait))
if DEVICE_CONNECTION_STRING and not args.suppress_messages:
send_iothub_message(iothub_client, msg_string)
time.sleep(args.wait)
async def main():
try:
if not sys.version >= "3.5.3":
raise Exception(
"The sample requires python 3.5.3+. Current version of Python: %s"
% sys.version)
print("IoT Hub Client for Python")
# The client object is used to interact with your Azure IoT hub.
module_client = IoTHubModuleClient.create_from_edge_environment()
# connect the client.
await module_client.connect()
prediction_client = PredictionClient(address='resnet50', port=50051)
classes_entries = requests.get(
"https://raw.githubusercontent.com/Lasagne/Recipes/master/examples/resnet50/imagenet_classes.txt"
).text.splitlines()
image_dir = "./assets/"
for image in os.listdir(image_dir):
try:
start_time = time.time()
results = prediction_client.score_file(
path=os.path.join(image_dir, image),
input_name='Placeholder:0',
outputs='classifier/resnet_v1_50/predictions/Softmax:0')
inference_time = (time.time() - start_time) * 1000
# map results [class_id] => [confidence]
results = enumerate(results)
# sort results by confidence
sorted_results = sorted(results,
key=lambda x: x[1],
reverse=True)
top_result = sorted_results[0]
msg_string = "(%.0f ms) The image %s was classified as %s with confidence %s." % (
inference_time, os.path.join(image_dir, image),
classes_entries[top_result[0]], top_result[1])
print(msg_string)
except:
print('except')
# define behavior for receiving an input message on input1
async def input1_listener(module_client):
while True:
input_message = await module_client.receive_message_on_input(
"input1") # blocking call
print("the data in the message received on input1 was ")
print(input_message.data)
print("custom properties are")
print(input_message.custom_properties)
print("forwarding mesage to output1")
await module_client.send_message_to_output(
input_message, "output1")
# define behavior for halting the application
def stdin_listener():
while True:
try:
selection = input("Press Q to quit\n")
if selection == "Q" or selection == "q":
print("Quitting...")
break
except:
time.sleep(10)
# Schedule task for C2D Listener
listeners = asyncio.gather(input1_listener(module_client))
print("The sample is now waiting for messages. ")
# Run the stdin listener in the event loop
loop = asyncio.get_event_loop()
user_finished = loop.run_in_executor(None, stdin_listener)
# Wait for user to indicate they are done listening for messages
await user_finished
# Cancel listening
listeners.cancel()
# Finally, disconnect
await module_client.disconnect()
except Exception as e:
print("Unexpected error %s " % e)
raise
async def processImages(module_client):
global scoring
global predictClient
global fps
async def scoreImage(module_client):
global scoring
global predictClient
global fps
currentFPS = fps.tick()[0]
image = cv2.imread("image.jpg")
result = predictClient.score_file(path="image.jpg",
input_name=input_tensor,
outputs=output_tensors)
classes, scores, bboxes = ssdvgg_utils.postprocess(
result, select_threshold=0.5)
if bboxes.size > 0:
message = {
"classes": str(classes.tolist()).strip(
'[]'), # because Azure IoT does not deal with list in JSON
"scores": str(scores.tolist()).strip('[]'),
"boxes": str(bboxes.tolist()).strip('[]'),
"datetime": str(datetime.datetime.now()),
"imageH": image.shape[0],
"imageW": image.shape[1],
"CameraURL": CameraURL,
"scoredFPS": currentFPS
}
await sendmessage(module_client, message)
scoring = False
while True:
try:
global predictClient
if FPGAinCloudAddress is not None and CameraURL is not None and FPGAinCloudaks_servicename is not None:
predictClient = PredictionClient(
address=FPGAinCloudAddress,
port=FPGAinCloudPort,
use_ssl=FPGAinCloudssl_enabled,
service_name=FPGAinCloudaks_servicename)
if "RTSP" in str(CameraURL).upper():
OpenCV = cv2.VideoCapture(CameraURL)
while True:
image = OpenCV.read()[1]
if scoring == False:
scoring = True
cv2.imwrite(image, "image.jpg")
scoreImage(module_client)
if PauseBetweenScoreSeconds > 0:
print("sleeping " + str(PauseBetweenScoreSeconds) +
" seconds between scoring")
time.sleep(PauseBetweenScoreSeconds)
elif "HTTP" in str(CameraURL).upper():
while True:
urllib.request.urlretrieve(url=CameraURL,
filename="image.jpg")
await scoreImage(module_client)
if PauseBetweenScoreSeconds > 0:
print("sleeping " + str(PauseBetweenScoreSeconds) +
" seconds between scoring")
time.sleep(PauseBetweenScoreSeconds)
else:
print(
"FPGAinCloudAddress and/or CamaraURL and/or FPGAinCloudaks_servicename in TWINS are not set."
)
time.sleep(15)
except Exception as e:
print(e)
print("sleeping for 15 seconds until retry")
time.sleep(15)
self.respond(bytes(json.dumps(message), "utf8"))
else:
self.respond(bytes("{}", "utf8"))
return
if __name__ == "__main__":
FPGAinCloudAddress = "40.121.68.174"
FPGAinCloudPort = 80
FPGAinCloudssl_enabled = False
FPGAinCloudaks_servicename = "fpga-aks-service"
if 'FPGAinCloudAddress' in os.environ.keys():
FPGAinCloudAddress = os.environ['FPGAinCloudAddress']
if 'FPGAinCloudPort' in os.environ.keys():
FPGAinCloudPort = os.environ['FPGAinCloudPort']
if 'FPGAinCloudssl_enabled' in os.environ.keys():
FPGAinCloudssl_enabled = util.strtobool(os.environ['FPGAinCloudssl_enabled'])
if 'FPGAinCloudaks_servicename' in os.environ.keys():
FPGAinCloudaks_servicename = os.environ['FPGAinCloudaks_servicename']
print("Connecting to " + FPGAinCloudAddress + ":" + str(FPGAinCloudPort) + " ssl:" + str(FPGAinCloudssl_enabled) + " service:" + FPGAinCloudaks_servicename)
# starting the PredictClient
predictClient = PredictionClient(address=FPGAinCloudAddress, port=FPGAinCloudPort,
use_ssl=FPGAinCloudssl_enabled, service_name=FPGAinCloudaks_servicename)
# starting the web server thread
_thread.start_new_thread(webServer, (80, ))
while True:
time.sleep(60) # allow the background thread to serve web request
def processImages(module_client):
global scoring
global predictClient
global fps
def scoreImage(module_client):
global scoring
global predictClient
global fps
currentFPS = fps.tick()[0]
image = cv2.imread("image.jpg")
result = predictClient.score_file(path="image.jpg",
input_name=input_tensor,
outputs=output_tensors)
classes, scores, bboxes = ssdvgg_utils.postprocess(
result, select_threshold=0.5)
if bboxes.size > 0:
message = {
"classes": classes,
"scores": scores,
"boxes": bboxes,
"datetime": str(datetime.datetime.now()),
"imageH": image.shape[0],
"imageW": image.shape[1],
"CameraURL": CameraURL,
"scoredFPS": currentFPS
}
IoTHubMessage = IoTHubMessage(bytearray(json.dumps(message)))
module_client.forward_event_to_output("output1", IoTHubMessage,
SEND_MESSAGECOUNTER)
scoring = False
while True:
try:
global predictClient
if FPGAinCloudAddress is not None and CameraURL is not None and FPGAinCloudaks_servicename is not None:
predictClient = PredictionClient(
address=FPGAinCloudAddress,
port=FPGAinCloudPort,
use_ssl=FPGAinCloudssl_enabled,
service_name=FPGAinCloudaks_servicename)
if "RTSP" in str(CameraURL).upper():
OpenCV = cv2.VideoCapture(CameraURL)
while True:
image = OpenCV.read()[1]
if scoring == False:
scoring = True
cv2.imwrite(image, "image.jpg")
scoreImage(module_client)
elif "HTTP" in str(CameraURL).upper():
while True:
urllib.request.urlretrieve(url=CameraURL,
filename="image.jpg")
scoreImage(module_client)
else:
print(
"FPGAinCloudAddress and/or CamaraURL and/or FPGAinCloudaks_servicename in TWINS are not set."
)
except Exception as e:
print(e)
print("sleeping for 15 seconds until retry")
time.sleep(15)