Convert TensorFlow model to ONNX
In the previous step of this tutorial, we created a machine learning model with TensorFlow. Now, we'll convert it to the ONNX format.
Here, we'll use the tf2onnx
tool to convert our model, following these steps.
- Save the tf model in preparation for ONNX conversion, by running the following command.
python save_model.py --weights ./data/yolov4.weights --output ./checkpoints/yolov4.tf --input_size 416 --model yolov4
- Install
tf2onnx
andonnxruntime
, by running the following commands.
pip install onnxruntime
pip install git+https://github.com/onnx/tensorflow-onnx
- Convert the model, by running the following command.
python -m tf2onnx.convert --saved-model ./checkpoints/yolov4.tf --output model.onnx --opset 11 --verbose
Next steps
We've now converted our model to an ONNX format, suitable for use with Windows Machine Learning APIs. In the final stage of this tutorial, we integrate it into a Windows app.