Convert torchvision models to OpenVino IR

[tominator]

Since the release of torchvision 0.5 all models in torchvision can be exported to ONNX for deployment. This includes models like Faster R-CNN and Mask R-CNN.

I want to convert them into the OpenVino IR but thats not possible. The export to ONNX requires opset version 11. But in OpenVino some operators just support opset version 10.

In the OpenVino Documentation I saw that they just support models from torchvision 0.2.  Are there any updates, when newer torchvision versions and models (like Faster R-CNN) will be supported? 

Thanks!

[Max_L_Intel · Accepted solution]

Hi @tominator 

There are some plans to support Mask RCNN and Faster RCNN models from PyTorch/ONNX, however, we can't really comment on this, since this might be a subject to change.

As for the opset-11 support, OpenVINO team is working on adding support for Resize-11. There are some PRs related to this already: operation specification and support for the operation in the Model Optimizer. 

Please keep an eye on OpenVINO updates related to new features.
Thank you.

[tominator]

Hi @Max_L_Intel ,

thanks for your reply.  I keep myself up to date.

Regards,

Tom

[Max_L_Intel]

If you need any additional information from Intel, please submit a new question as this thread is no longer being monitored.