predict.py

import argparse
import tensorflow as tf
tf.get_logger().setLevel(3)

from scipy.io import loadmat, savemat

parser = argparse.ArgumentParser(description='Perform inference with a Neural Network by a mat file')
parser.add_argument('--input', default='input.mat', help='input mat file path')
parser.add_argument('--output', default='output.mat', help='output mat file path')
parser.add_argument('--model', default='mymodel', help='tf savedmodel directory')

if __name__=='__main__':
    args = parser.parse_args()
    print('Loading model')
    model = tf.keras.models.load_model(args.model)
    print(f'Model inputs: {model.inputs}')
    print(f'Model outputs: {model.outputs}')

    print(f'Loading input file {args.input}')
    inputs = loadmat(args.input)['x']

    print(f'Invoking model...')
    outputs = model.predict(inputs)

    print(f'Saving output file {args.output}')
    savemat(args.output, {
        'y': outputs
    })    

requirements.txt

tensorflow
scipy

TF_MATLAB_RUNNER.md

Setup and Requirements

• A tensorflow supported python version (3.6-3.8)
• (Optionally) CUDA Drivers for your nvidia GPU
• A TF SavedModel to use
• Run pip install -r requirements.txt to install required packages (tensorflow, scipy)
• A mat file with an array of proper size named 'x'

Sample Usage

Assuming that you run python in a directory like:

<your directory>
│   predict.py
│   README.md
│   requirements.txt
│   input.mat
│
└───mymodel
    │   saved_model.pb
    │
    ├───assets
    └───variables
            variables.data-00000-of-00001
            variables.index

python predict.py --input sample_input.mat --output output.mat --model mymodel Absolute paths are also supported.