How to write custom layers extensions for ONNx Models

[Raza__Ghulam_Jilani]

If you have a model which you translated from Pytorch to ONNX, and has custom layers in  it, can you write custom layer extensions for model optimizer and IE in the same way like TF,Caffe and MXNET? becasue there is no mention of this in OpenVino guides, so had to ask.

Thanks ,

Raza

[JesusE_Intel]

Hi Raza,

Thanks for reaching out. It should be possible to write custom layer extensions with ONNX similar to TF/Caffe. We are working on updating the documentation, I will see if I can get a time estimate. 

Could you share some information about your use case and I will provide feedback to the team? Let me know if you would like to send this privately and I will start a private message.

Regards,

Jesus

[Raza__Ghulam_Jilani]

Hi Jesus,

Thanks for reaching out. Actually I am trying to convert my own implementation of YOLO3 from pytorch to IR format. One way i read in the docs was to convert it to onnx first and then to IR. But there is no doc for writing custom layer extensions for onnx and also if you could add a tutorial for converting custom pytorch models, that would be great.

Raza

[Muaz__Urwa]

Any updates on this ?