Train suite of models with Tensorflow to be benchmarked about tf-encrypted

[justin1121]

A trained vgg16 model can be exported like this:

import tensorflow.keras as K

from tensorflow.python.framework import graph_util
from tensorflow.python.framework import graph_io
import tensorflow as tf

vgg16 = tf.keras.applications.VGG16()

sess = K.backend.get_session()

num_output = 1
pred = [None]*num_output
pred_node_names = [None]*num_output
for i in range(num_output):
    pred_node_names[i] = "hi"+str(i)
    pred[i] = tf.identity(vgg16.outputs[i], name=pred_node_names[i])
print('output nodes names are: ', pred_node_names)

constant_graph = graph_util.convert_variables_to_constants(sess,
                                   sess.graph.as_graph_def(),
                                   pred_node_names)

frozen = graph_util.remove_training_nodes(constant_graph)

output = "vgg16.pb"
graph_io.write_graph(frozen, ".", output, as_text=False)
print('saved the freezed graph (ready for inference) at: ', output)

This might be able to be done with other models from keras.

[yanndupis]

Great! there was also this post where they convert .h5 file to protobuf file: https://medium.com/@johnsondsouza23/export-keras-model-to-protobuf-for-tensorflow-serving-101ad6c65142