Error in importing trained TensorFlow SavedModel using importTensorFlowNetwork function
21 views (last 30 days)
Iman Askari on 8 Sep 2021
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:
net = importTensorFlowNetwork('EXP1_pb_model');
The python code I used to convert the ckpt to pb is as follow:
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')
# Export checkpoint to SavedModel
builder = tf.compat.v1.saved_model.builder.SavedModelBuilder(export_dir)
Attached are the both the ckpt and pb models.
I would appericiate any help in resolving this issue.
Anshika Chaurasia on 14 Sep 2021
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!