Package Installation

This lesson will illustrate how to install Python dependencies in our virtual environment.

We’ll need the following Python packages to export our models into:

• ONNX models
• TensorFlow models in SavedModel format
• TFLite models

PyTorch

We can easily install the stable version of PyTorch as follows:

pip install torch

TensorFlow

To install TensorFlow, use the following command:

pip install tensorflow

ONNX

The onnx package is useful for verifying the underlying protocol buffer (protobuf) of our model. Install it as follows:

pip install onnx

ONNXRuntime

For inference using the ONNX model, we’ll need the onnxruntime package. Run the following command to install it:

pip install onnxruntime

ONNX-TensorFlow

We cannot directly convert the ONNX model into a TFLite model. The onnx-tensorflow is one of the TensorFlow converter projects. It provides the TensorFlow backend for ONNX.

Install it as follows:

git clone https://github.com/onnx/onnx-tensorflow.git && cd onnx-tensorflow
pip install -e .