ONNX (Open Neural Network Exchange) is a standard that acts as a universal translator for deep learning models. It enables models that were trained in one framework (TensorFlow, PyTorch, etc.) to be run in ENVI Deep Learning, without the need to rewrite the code.

An example workflow follows:

• Use ONNX tools to export (convert) models from other frameworks to ONNX models.
• Use the ENVI Deep Learning Configure ONNX Model tool configures the ONNX models to run in ENVI.
• Use the *.envi.model files created with Configure ONNX Model to run ENVI Deep Learning classification tasks. ONNX Runtime, used for inference, leverages hardware accelerators to increase inference speed.

Other ONNX tools in ENVI Deep Learning include the following:

• Edit ONNX Training Model Metadata, to edit metadata that was saved with the trained model file.
• Publish ENVI ONNX Model, to share Deep Learning ONNX models with other users on the ENVI Repository.

Frameworks that export models to ONNX format include:

• PyTorch
• TensorFlow
• Keras
• Scikit-learn
• XGBoost
• LightGBM
• MXNet
• Chainer
• CoreML
• PaddlePaddle
• MATLAB
• CatBoost

For additional details about ONNX, see the ONNX website.

See Also

---

Configure ONNX Model, Publish ENVI ONNX Model, Edit ONNX Training Model Metadata