TensorFlow MNIST

Show saving a simple model and using the saved model for inference.

Train model

To train a basic Softmax regression model, do the following:

python ./train_basic_model.py

To run a one-convolutional layer model, do the following:

python ./train_cnn_model.py

When the script completes, the exported model we should use will be saved, by default, as ./saved-model/optimized_mnist_model.pb. This is a Protocol Buffer file which represents a serialized version of our model with all the learned weights and biases. This specific one is optimized for inference-only usage.

Use model

Use the following script to classify an image of a digit. The image should be a 28x28 image with the background black and foreground white.

python ./classify_mnist.py <img path>

Optional flags for this are:

• --graph-file for specifying your saved model file. Default is ./saved-model/optimized_mnist_model.pb.

Several sample images to use for inference are in the sample-images directory of the project.