incorrect parameter with Conv3DTranspose
ahwiggers opened this issue · 2 comments
I'm getting the following error when trying to run inference using an ONNX model using the WindowsML package in a UWP app.
'The parameter is incorrect.
Non-zero status code returned while running Node: StatefulPartitionedCall/model/conv3d_transpose/conv3d_transpose Status Message: Input X must be 4-dimensional. X: {1,256,16,16,8}'
The model is using ONNX opset 9 and was converted from a tensorflow model using tf2onnx versions 1.12.1 and I'm running in VisualStudio 2019. I saw a resolved issue from 2019 that stated it wasn't supported (but would be soon), so I was wondering if it's possible this layer is still not supported by WindowsML? If it's not supported (and isn't intended to be anytime soon), is there any updated guidance on how to make a custom layer?
Thanks for posting. Are you using the inbox or NuGet version of WinML? The NuGet version should have support for conv3d_transpose
Closing due to inactivity.