TFLite Model Maker

Overview

The TFLite Model Maker library simplifies the process of adapting and converting a TensorFlow neural-network model to particular input data when deploying this model for on-device ML applications.

Requirements

Refer to requirements.txt for dependent libraries that're needed to use the library and run the demo code.
Note that you might also need to install sndfile for Audio tasks. On Debian/Ubuntu, you can do so by sudo apt-get install libsndfile1

Installation

There are two ways to install Model Maker.

Install a prebuilt pip package: tflite-model-maker.

pip install tflite-model-maker

If you want to install nightly version tflite-model-maker-nightly, please follow the command:

pip install tflite-model-maker-nightly

Clone the source code from GitHub and install.

git clone https://github.com/tensorflow/examples
cd examples/tensorflow_examples/lite/model_maker/pip_package
pip install -e .

TensorFlow Lite Model Maker depends on TensorFlow pip package. For GPU support, please refer to TensorFlow's GPU guide or installation guide.

End-to-End Example

For instance, it could have an end-to-end image classification example that utilizes this library with just 4 lines of code, each of which representing one step of the overall process. For more detail, you could refer to Colab for image classification.

Step 1. Import the required modules.

from tflite_model_maker import image_classifier
from tflite_model_maker.image_classifier import DataLoader

Step 2. Load input data specific to an on-device ML app.

data = DataLoader.from_folder('flower_photos/')

Step 3. Customize the TensorFlow model.

model = image_classifier.create(data)

Step 4. Evaluate the model.

loss, accuracy = model.evaluate()

Step 5. Export to Tensorflow Lite model and label file in export_dir.

model.export(export_dir='/tmp/')

Notebook

Currently, we support image classification, text classification and question answer tasks. Meanwhile, we provide demo code for each of them in demo folder.