Error in importing trained TensorFlow SavedModel using importTensorFlowNetwork function

I am trying to import a trained tensoflow neural network model. Initially the trained model is in checkpoint format (ckpt). I was able to convert the ckpt to savedModel (pb) format for use in importTensorFlowNetwork function. While running the function I obtain the following error:

>>

unzip('EXP1_pb_model.zip') 
net = importTensorFlowNetwork('EXP1_pb_model');

Importing the saved model...

Unrecognized field name "object_graph_def".

Error in nnet.internal.cnn.tensorflow.savedmodel.TFSavedModel (line 32)
obj.KerasManager = savedmodel.TFKerasManager(smstruct.meta_graphs.object_graph_def, obj.SavedModelPath, kerasImporterOptions, importNetwork);

Error in nnet.internal.cnn.tensorflow.importTensorFlowNetwork (line 21)
sm = savedmodel.TFSavedModel(path, options, true);

Error in importTensorFlowNetwork (line 107)
Network = nnet.internal.cnn.tensorflow.importTensorFlowNetwork(modelFolder, varargin{:});

The python code I used to convert the ckpt to pb is as follow:

import os

import tensorflow as tf

trained_checkpoint_prefix = '/exp_1/model.ckpt'

export_dir = os.path.join('model', 'EXP1_model')

graph = tf.Graph()

with tf.compat.v1.Session(graph=graph) as sess:

# Restore from checkpoint

loader = tf.compat.v1.train.import_meta_graph(trained_checkpoint_prefix + '.meta')

loader.restore(sess, trained_checkpoint_prefix)

# Export checkpoint to SavedModel

builder = tf.compat.v1.saved_model.builder.SavedModelBuilder(export_dir)

builder.add_meta_graph_and_variables(sess,

[tf.saved_model.TRAINING, tf.saved_model.SERVING],

strip_default_attrs=True)

builder.save()

Attached are the both the ckpt and pb models.

I would appericiate any help in resolving this issue.

Thanks

Answer 1

[Anshika Chaurasia]

Hi Iman,

We currently support the import of TF models saved using the Sequential and Funtional Keras Model APIs (https://keras.io/guides/sequential_model/ & https://keras.io/guides/functional_api/ ) . But the models in your case do not use that API to save the model.

The importTensorFlowNetwork look for a graph called "object_graph_def" in the saved model and does not find it if the model was saved as it is in your case (hence the "Unrecognized field" error).

Hope it helps!