Deploy-Models-with-TensorFlow-Serving-and-Flask

What's Included

Following folders and files are included in this repo:

1. pets - TensorFlow SavedModel Directory
2. static - Empty directory which will be used for storing images by the flask app
3. templates - HTML templates are here
4. app.py - Flask app
5. README.md - This file

Environment

You will require Python3 installed. I used python 3.7 and TensorFlow 2.1.0, and I'd recommend you do the same. It is recommended that you create a new virtual environment to avoid issues with existing installations.

Install the python packages required:

$ pip3 install tensorflow==2.1.0 flask flask-bootstrap requests

Docker Instance

Launch the docker instance which will serve the TensorFlow SavedModel (in the pets folder):

$ docker run -p PORT_NUMBER:8501 --name=pets -v "YOUR_SAVED_MODEL_PATH:/models/pets/1" -e MODEL_NAME=pets tensorflow/serving

In the project, I've used 8502 for the PORT_NUMBER , and YOUR_SAVED_MODEL_PATH needs to be the absolute path of the pets folder in your local machine. So, if you extracted the downloaded zip file in, say, /home/example/ , and want to use 8502 for the server port, the above command will become:

$ docker run -p 8502:8501 --name=pets -v "/home/example/pets/:/models/pets/1" -e MODEL_NAME=pets tensorflow/serving

Please note if you use any other port, you will have to change the MODEL_URI in the app.py file accordingly.

Flask App

Once the docker instance is running, you can launch the flask app:

$ python3 app.py

And that's it! The default port for flask is 5000, so you can access the app by going to localhost:5000 in your browser.