Steps needed to export Keras model .h5 file to .pb file used by Tensorboard in order to display network graph.

Keras2TensorBoard.py

# How to use this:
#
# Step 1. You need to export Keras model to .h5 file
# which includes network architecture:
# https://github.com/amir-abdi/keras_to_tensorflow

# You can do this by including (**) line below your model definition
# for example, in my case:
#        inception_v3 = keras.applications.inception_v3.InceptionV3(
#            include_top=True,
#            weights=None,
#            input_tensor=None,
#            input_shape=img_shape,
#            pooling=None,
#            classes=num_classes)
#
#        inception_v3.load_weights(pretrained_model, by_name=True)

#        # (**) needed this to export model architecture.
#        inception_v3.save('inception_v3_saved_model_mnedic.h5')

# Step 2.
#   $ python3 keras_to_tensorflow.py -input_model_file inception_v3_saved_model_mnedic.h5
#   or
#   $ python keras_to_tensorflow.py -input_model_file inception_v3_saved_model_mnedic.h5

# This should generate:
#   inception_v3_saved_model_mnedic.h5.pb file

# Step 3. Run: $ python3 tensorboard_graph.py
# Step 4. Run: $ tensorboard --logdir=path/to/log-directory/
#                tensorboard --logdir=logs/tests/2/

# Step 5. Navigate your browser to shown address. You should see graph

import tensorflow as tf
from tensorflow.python.platform import gfile

with tf.Session() as sess:
    model_filename = '/home/mnedic/int_reload/inception_v3_saved_model_mnedic.h5.pb'
    with gfile.FastGFile(model_filename, 'rb') as f:
        graph_def = tf.GraphDef()
        graph_def.ParseFromString(f.read())
        g_in = tf.import_graph_def(graph_def)
LOGDIR = 'logs/tests/1/'
train_writer = tf.summary.FileWriter(LOGDIR)
train_writer.add_graph(sess.graph)

train_writer.flush()
train_writer.close()