onnx/onnx-tensorflow

What it solves

This project provides a way to use machine learning models created in the ONNX (Open Neural Network Exchange) format, allowing them to be executed within the TensorFlow ecosystem. It solves the problem of framework interoperability, enabling users to migrate models from other frameworks that export to ONNX into TensorFlow for inference.

How it works

The tool converts an ONNX model into a TensorFlow model representation. Once converted, the model is delegated to TensorFlow for execution to produce the final output. This can be done either through a command-line interface (CLI) or programmatically via Python code.

Who it’s for

Developers and machine learning engineers who have models in ONNX format and need to run them using TensorFlow's execution engine and hardware optimizations.

Highlights

• own-line interface for quick model conversion (onnx-tf convert).
• Support for programmatic conversion and direct inference using the prepare function.
• Integration with the ONNX Model Zoo to verify conversion success across a variety of models.
• Clear distinction from tf2onnx, which performs the opposite operation (TensorFlow to ONNX).