error in readNetFromTensorflow in c++

Question

I'm new in deep learning. In first step I create and train a model in python with keras and freezed by this code:

def export_model(MODEL_NAME, input_node_name, output_node_name):

    tf.train.write_graph(K.get_session().graph_def, 'out', \
        MODEL_NAME + '_graph.pbtxt')

    tf.train.Saver().save(K.get_session(), 'out/' + MODEL_NAME + '.chkp')

    freeze_graph.freeze_graph('out/' + MODEL_NAME + '_graph.pbtxt', None, \
        False, 'out/' + MODEL_NAME + '.chkp', output_node_name, \
        "save/restore_all", "save/Const:0", \
        'out/frozen_' + MODEL_NAME + '.pb', True, "")

    input_graph_def = tf.GraphDef()
    with tf.gfile.Open('out/frozen_' + MODEL_NAME + '.pb', "rb") as f:
        input_graph_def.ParseFromString(f.read())

    output_graph_def = optimize_for_inference_lib.optimize_for_inference(
            input_graph_def, [input_node_name], [output_node_name],
            tf.float32.as_datatype_enum)

    with tf.gfile.FastGFile('out/opt_' + MODEL_NAME + '.pb', "wb") as f:
        f.write(output_graph_def.SerializeToString())

it's output :

• checkpoint
• Model.chkp.data-00000-of-00001
• Model.chkp.index
• Model.chkp.meta
• Model_graph.pbtxt
• frozen_Model.pb
• opt_Model.pb

when I want to read the net in opencv c++ by readNetFromTensorflow :

String weights = "frozen_Model.pb";
String pbtxt = "Model_graph.pbtxt";
dnn::Net cvNet = cv::dnn::readNetFromTensorflow(weights, pbtxt);

This will make error :

OpenCV(4.0.0-pre) Error: Unspecified error (FAILED: ReadProtoFromBinaryFile(param_file, param). Failed to parse GraphDef file: frozen_Model.pb) in cv::dnn::ReadTFNetParamsFromBinaryFileOrDie, file D:\LIBS\OpenCV-4.00\modules\dnn\src\tensorflow\tf_io.cpp, line 44

and

OpenCV(4.0.0-pre) Error: Assertion failed (const_layers.insert(std::make_pair(name, li)).second) in cv::dnn::experimental_dnn_v4::`anonymous-namespace'::addConstNodes, file D:\LIBS\OpenCV-4.00\modules\dnn\src\tensorflow\tf_importer.cpp, line 555

How to fix this error?

Answer 1

Amin, may I ask you to try to save a graph in a testing mode:

K.backend.set_learning_phase(0)  # <--- This setting makes all the following layers work in test mode

model = Sequential(name = MODEL_NAME)
model.add(Conv2D(filters = 128, kernel_size = (5, 5), activation = 'relu',name = 'FirstLayerConv2D_No1',input_shape = (Width, Height, image_channel)))
...
model.add(Dropout(0.25))
model.add(Dense(100, activation = 'softmax', name = 'endNode'))

# Create a graph definition (with no weights)
sess = K.backend.get_session()
sess.as_default()
tf.train.write_graph(sess.graph.as_graph_def(), "", 'graph_def.pb', as_text=False)

Then freeze your checkpoint files with a newly created graph_def.pb by a freeze_graph.py script (do not forget to use --input_binary flag).