opencv: Import tensorflow pre-trained model using dnn modules encount an error

System information (version)

• OpenCV => 3.3.0
• Operating System / Platform => Windows 10(64 Bit)
• Compiler => Visual Studio 2015
• Tensorflow => 1.3.0
• Python => 3.6.1

Detailed description

I’ve wrote a lenet-5 model using tensorflow to train the mnist dataset, then I save the trained model file in .pb format. I use the dnn::readNetFromTensorflow() function to read the .pb file, but I get the following error:

conv1/W_conv1/RandomStandardNormal:RandomStandardNormal(conv1/W_conv1/shape)
seed:[ ]
seed2:[ ]
dtype:0
T:0
OpenCV Error: Unspecified error (Unknown layer type RandomStandardNormal in op conv1/W_conv1/RandomStandardNormal) in cv::dnn::experimental_dnn_v1::`anonymous-namespace'::TFImporter::populateNet, file C:\build\master_winpack-build-win64-vc14\opencv\modules\dnn\src\tensorflow\tf_importer.cpp, line 883

Here are my tensorflow code of “conv1/W_conv1” part:

def weight_variable(shape, name):
    initial = tf.random_normal(shape, stddev=0.1, name=name)
    return tf.Variable(initial)
# conv1 layer
with tf.name_scope("conv1"):
    W_conv1 = weight_variable([5, 5, 1, 32], name='W_conv1') 
    b_conv1 = bias_variable([32], name='b_conv1')
    h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1, name='h_conv1') 

I couldn’t find any problem in my tensorflow code, so I guess maybe my tensorflow version(released in Aug 17) is too updated for opencv 3.3.0(released in Aug 3). Hope anybody who can help me with this, thanks in advance.