CODEWITHSUNDEEP
pythontensorflow
user2867237
user2867237

Reputation: 457

Switch between graphs in tensorflow

I am working on a project that involves face detection, face recognition (based on facenet), age/gender detection, and face expression analysis. For each mentioned feature, I have one tensorflow graph that works fine. Now, I need to combine all of them in one single code. My approach is as follow:

with tf.Graph().as_default(): 

      sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
      with sess.as_default():

            #configure my camera
            video_capture = cv2.VideoCapture(0)


            while True:

              #read one frame from camera
              ret, frame = video_capture.read()



             #here I load face recognition graph using saver restore
             facenet.load_model(modeldir)
             images_placeholder tf.get_default_graph().get_tensor_by_name("input:0")
             embeddings = tf.get_default_graph().get_tensor_by_name("embeddings:0")


             #Here I reset the graph and load another tensorflow graph for age/gender detection
             tf.reset_default_graph()
             ....


             #Here I reset the graph and load another tensorflow graph for face expression analysis
             tf.reset_default_graph()
             .....

Now my issue is that the code is not efficient and very slow. The reason is that for each frame of the video, I need to load (restore) several graphs from my disk (in while). But, I would like to load all the graphs once (before while), and just switch the graphs in the while loop to reduce the runtime. I would appreciate your comment

Upvotes: 2

Views: 1303

Answers (1)

Sraw
Sraw

Reputation: 20214

Try the following form:

graph_face_detection = tf.Graph()
sess_face_detection = tf.Session(graph=graph_face_detection)

graph_face_recognition = tf.Graph()
sess_face_recognition = tf.Session(graph=graph_face_recognition)

...

sess_face_detection.run(operations)

...

sess_face_recognition.run(operations)

If you want to use gpu, be careful about your gpu memory allocation.

Update

First, one session one graph.

Second, to specify which graph operation should use:

graph_face_detection = tf.Graph()
sess_face_detection = tf.Session(graph=graph_face_detection)

graph_face_recognition = tf.Graph()
sess_face_recognition = tf.Session(graph=graph_face_recognition)

with graph_face_detection.as_default() as g:
    output_face_detection = tf.matmul(x, y)

with graph_face_recognition.as_default() as g:
    output_face_recognition = tf.matmul(x, y)

...

sess_face_detection.run([output_face_detection])
sess_face_recognition.run([output_face_recognition])

Further, when you run output_face_detection = tf.matmul(x, y), you just create a node and add it into a graph without any actual calculation. So you can build all your graphs first, and then use sess_xxx.run(operation) in the loop.

Upvotes: 2

Related Questions