如何部署使用 export_saved_model 保存的 TensorFlow 模型

Question

I have a.pbtxt file that I have through export_saved_model from inception_v4, however I can't use this saved_model to do prediction. When I try to load the model using tf.contrib.predictor.from_saved_model() I get the following error:

OSError: Cannot parse file b'/Users/mehdi/Desktop/serving/saved_model.pbtxt': 1:1 :   
Message type "tensorflow.SavedModel" has no field named "node"..

Answer 1

You can do it using two ways:

1. By Loading the Saved Model and Predicting using the Loaded Model

   New_Model = tf.keras.models.load_model("saved_model")

   New_Model.summary()

   Prediction = New_Model.predict(...)

   2. Using Tensorflow Serving: Executing the below commands in the Terminal

Install the Tensorflow Serving using the Docker Image

sudo docker pull tensorflow/serving

# Invoke the Tensorflow Model Server
sudo docker run -p 8501:8501 --mount type=bind,source=Path_Of_The_Saved_Model_In_PC,target=/models/saved_model -e MODEL_NAME=saved_model -t tensorflow/serving &

#To get the status of the model
curl http://localhost:8501/v1/models/saved_model

curl -d '{"instances": [1.0, 2.0, 5.0]}' \
    -X POST http://localhost:8501/v1/models/saved_model:predict

If you have an Image as Input and if you want to perform some pre-processing of that image in a Client File, you can do it using the commands mentioned below:

sudo docker pull tensorflow/serving

sudo docker run -p 8501:8501 --mount type=bind,source=Path_Of_The_Model,target=/models/saved_model -e MODEL_NAME=saved_model -t tensorflow/serving &

python Path/client.py --num_tests=100 --server=localhost:8500

Please refer this link for Code for Client file for TF Serving, this Guide for more info on TF Serving and this Tutorial for End to End Example.