How to apply ONNX models to TAO Toolkit?

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

I’d like to apply ONNX models to TAO Toolkit.
I found TAO BYOM Converter, however the document says that BYOM supports only Classification and UNet and does not support others like Object Detection.
If I use ONNX model with TAO, am I supposed to use TAO BYOM Converter TF2?
I read the document and Jupyter Notebooks of TAO Toolkit, but I could not reach how to and another way except using TAO BYOM Converter.
And I could not find any Jupyter Notebook this video says.
Would you mind letting me know the specific way to apply ONNX model to TAO?

Ubuntu: 22.04
TAO Toolkit: 5.3

2

Yes, for 3rd-party onnx file, currently only Classification and UNet are supported.

There is an example in GitHub - NVIDIA-AI-IOT/tao_byom_examples: Examples of converting different open-source deep learning models to TAO compatible format through TAO BYOM package..

3

@Morganh
Thank you for your reply.

Yes, for 3rd-party onnx file, currently only Classification and UNet are supported.

I see, so I cannot use ONNX models with TAO Toolkit if they are not the models of Classification and UNet.

There is an example in GitHub - NVIDIA-AI-IOT/tao_byom_examples: Examples of converting different open-source deep learning models to TAO compatible format through TAO BYOM package..

I had read the GitHub page, but I could not understand apparently.
So, is all I have to do moving .tltb files into $LOCAL_EXPERIMENT_DIR/pretrained_resnet18/ and run the Jupyter Notebook?

And is the web page mainteined?
I clicked several url links, but most of them were 404 pages.

4

Please refer to the latest 5.3.0 notebook.

5

Please refer to the latest 5.3.0 notebook.

Thank you so much.

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.