Since Tensorflow 2 Alpha was annonced recently, I wanted to try it out by making a simple image classification model trained on fashion mnist data set using a 3 layers neural networks. The model is served as a Rest API.

Defined classes are ‘T-shirt/top’, ‘Trouser’, ‘Pullover’, ‘Dress’, ‘Coat’, ‘Sandal’, ‘Shirt’, ‘Sneaker’, ‘Bag’, ‘Ankle boot’.

Training set

keras.datasets.fashion_mnist

Libs

• Tensorflow 2.0.0-alpha0
• Keras
• Python 3
• skimage
• Flask
• Numpy

How to run?

1- Install the required libs (with conda install or pip install).

2- Run python train_and_export_model.py to train the deep neural networks and export to export it to fashion_mnist_classifier savedModel.

3- Run python app.py to start the server that runs the port 5000.

4- Send a POST query to the http://127.0.0.1/api/recognize_image with a body consisting of a JSON content:

   {
      "img_url" : "<IMG_TO_CLASSIFY_URL> (http://www.stickpng.com/assets/thumbs/580b57fbd9996e24bc43bef3.png)"
    }

And the Server should do the prediction and then respond with :

  {
    "data": {
    "confidence": "89%",
    "prediction": "Dress"
  },
    "isError": false,
    "message": "Success",
    "statusCode": 200
  }