Facenet: Error while generating embeddings

[gengstah]

Hello,

I got this error while generating embeddings:

Exception in thread Thread-1:
Traceback (most recent call last):
  File "C:\Users\gengstah\Anaconda3\envs\facenet\lib\threading.py", line 916, in _bootstrap_inner
    self.run()
  File "C:\Users\gengstah\Anaconda3\envs\facenet\lib\threading.py", line 864, in run
    self._target(*self._args, **self._kwargs)
  File "contributed\face-recognition.py", line 84, in capture
    faces = self.face_recognition.identify(frame)
  File "C:\Users\gengstah\python-workspace\facenet\contributed\face.py", line 80, in identify
    face.embedding = self.encoder.generate_embedding(face)
  File "C:\Users\gengstah\python-workspace\facenet\contributed\face.py", line 110, in generate_embedding
    images_placeholder = tf.get_default_graph().get_tensor_by_name("input:0")
  File "C:\Users\gengstah\Anaconda3\envs\facenet\lib\site-packages\tensorflow\python\framework\ops.py", line 2734, in get_tensor_by_name
    return self.as_graph_element(name, allow_tensor=True, allow_operation=False)
  File "C:\Users\gengstah\Anaconda3\envs\facenet\lib\site-packages\tensorflow\python\framework\ops.py", line 2584, in as_graph_element
    return self._as_graph_element_locked(obj, allow_tensor, allow_operation)
  File "C:\Users\gengstah\Anaconda3\envs\facenet\lib\site-packages\tensorflow\python\framework\ops.py", line 2627, in _as_graph_element_locked
    "graph." % (repr(name), repr(op_name)))
KeyError: "The name 'input:0' refers to a Tensor which does not exist. The operation, 'input', does not exist in the graph."

I rewrote the contributed/real_time_face_recognition.py to use tkinter. Since the continuous loop of grabbing an image from camera, predicting the face, and displaying it in a tk.Label (tkinter) is causing the application not to be responsive, I decided to implement these processes in a separate thread. After doing so, I get the error above. Here is a snippet of what I have done so far:

    def __init__(self, master=None):
        super().__init__(master)
        print("Loading Face Recognition...")
        self.face_recognition = face.Recognition()
        print("Done")

    def start(self):
        print("start")
        self.stop_event = threading.Event()

        self.channel1 = threading.Thread(target=self.capture, args=("0", self.face_recognition))
        self.channel1.start()

    def capture(self, channel, face_recognition):
        frame_interval = 3  # Number of frames after which to run face detection
        fps_display_interval = 5  # seconds
        frame_rate = 0
        frame_count = 0

        start_time = time.time()

        while(not self.stop_event.is_set()):
            try:
                url_response = urllib.request.urlopen("[SNAPSHOT_IMAGE_FROM_IP_CAMERA]")
            except Exception as e:
                print("Failed to get feed from channel {}: {}".format(channel, e))
                break
            img_array = np.array(bytearray(url_response.read()), dtype=np.uint8)
            frame = cv2.imdecode(img_array, -1)

            # Identify faces
            if (frame_count % frame_interval) == 0:
                faces = face_recognition.identify(frame)

                # Check our current fps
                end_time = time.time()
                if (end_time - start_time) > fps_display_interval:
                    frame_rate = int(frame_count / (end_time - start_time))
                    start_time = time.time()
                    frame_count = 0

            self.add_overlays(frame, faces, frame_rate)

            frame_count += 1

            b,g,r = cv2.split(frame)
            frame = cv2.merge((r,g,b))
            img = Image.fromarray(frame)
            imagetk = ImageTk.PhotoImage(image=img)

            if channel == "0":
                self.imageview1.configure(image=imagetk)
                self.imageview1.image = imagetk

EDIT 1: I am using the frozen graph provided by David and generated my own classifier. The model and classifier is working nicely using my webcam.

EDIT 2: I tried to store the face.Recognition object to thread local and tested it in one thread. It works but having more than 1 thread (more than 1 face.Recognition object) eats up all my memory. How can I use only one face.Recognition for many threads without getting the error above?

[gengstah]

Ok, I got it working by modifying face.py. I learned that tf.get_default_graph() returns a Graph for the current thread. This works fine in a single-threaded application but since I'm working on a multi-threaded application, the tf.get_default_graph() returns a Graph that doesn't have my model loaded.

I changed the following lines:

images_placeholder = tf.get_default_graph().get_tensor_by_name("input:0")
embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")
phase_train_placeholder = tf.get_default_graph().get_tensor_by_name("phase_train:0")

to

images_placeholder = self.sess.graph.get_tensor_by_name("input:0")
embeddings = self.sess.graph.get_tensor_by_name("embeddings:0")
phase_train_placeholder = self.sess.graph.get_tensor_by_name("phase_train:0")

This change enables multi-threading.