[ST Edge AI Core v4.0.1] Converting ONNX model with a dynamic Convolution layer

[Mariusz_R]

I’m trying to convert an ONNX model which has a Conv layer that takes weights as input (has 2 inputs). Edge AI gives me:

NOT IMPLEMENTED: Convolution with multiple not constant inputs not supported

Are there any short-term plans to support such layers in ST Edge AI?

I also tried to change the name of the layer to some custom name (DynConv). The error in ST Edge AI changed to:

NOT IMPLEMENTED: Unsupported layer types: DynConv

I thought that I could bypass this error by using --custom with a JSON definition, but the JSON format is poorly documented. Also, many attempts at different formats seem to indicate that --custom only works for Keras. Is this correct?

[Julian E.]

Hi ​@Mariusz_R,

 

Indeed dynamic convolution layers are not supported.

I did not hear of any plan to supported them soon.

This could change if a big customer request it. If you work for a big company, please reach out to your local FAE.

If not, sadly, you will have to wait.

 

We are doing a lot of things regarding AI regarding both Software and Hardware but we cannot do as much as we would like, this is currently not a priority.

 

For customer layers you are correct. It is badly documented (but not very much used) and only for keras.

My understanding is that it is much more difficult to handle for ONNX as this is a completely different thing.

This is not planned to support it for ONNX.

 

Have a good day,

Julian