I’ve been investigating the compatibility between Maya’s ML Deformer and Houdini’s ML Deformer.

Since both tools export ONNX files and perform inference from them, I'm wondering whether they could share a common ONNX model.

In my test, I used orientation information as an input feature in the dataset. However, through the training, I noticed that the tensor shape was 1-dimensional in Maya and 2-dimensional in Houdini, as shown below:
Maya’s input shape:
Houdini’s input shape:
(*This was confirmed using the .ipynb script I’ve attached.)

This means the input structure of the ONNX files generated by each tool differs, and an ONNX file created in Houdini cannot be directly used in Maya.

In Houdini, I used the ML Regression Train TOP node to generate the ONNX file without writing any code.

My question is:
Is it possible to configure this node to output 1-dimensional tensors without any coding(to match Maya’s expected input)?

Additionally, if anyone has insights or experience regarding interoperability between Houdini’s and Maya’s ML Deformer systems, I’d really appreciate your input.


Environment:
Houdini 20.5.445
Maya 2025.3

Sorry, I attached the wrong hip file.
Here is the correct one.