Not able to use dynamic batching on mask rcnn model

[Ramasabarinathan]

Hello,

         I referred https://docs.openvinotoolkit.org/latest/openvino_docs_IE_DG_DynamicBatching.html and trying to implement the dynamic batching with the /intel/openvino/deployment_tools/open_model_zoo/models/public/mask_rcnn_inception_v2_coco. I have converted the model into IR and used to implement the dynamic batching. But when loading the network in python using (ie.load_network(network=net, device_name=device, num_requests=num_reqs)), it ends in error as below.

exec_net = ie.load_network(network=net, device_name=device, num_requests=num_reqs)
File "ie_api.pyx", line 336, in openvino.inference_engine.ie_api.IECore.load_network
File "ie_api.pyx", line 345, in openvino.inference_engine.ie_api.IECore.load_network
RuntimeError: MKLDNNGraph::CreateGraph: such topology cannot be compiled for dynamic batch!

But the same code is working in Image classification models like GoogleNet, AlexNet but i need to use the mask rccn model.

Please let me know how to make the mask rcnn model to work with dynamic batching.

[IntelSupport]

Hi Ramasabarinathan,

Thanks for reaching out.

According to your error, I suspect your model must have unsupported layers by the dynamic batching feature. Check out the list of supported layers from this documentation:

https://docs.openvinotoolkit.org/latest/openvino_docs_IE_DG_DynamicBatching.html

If possible, you could share your model for us to verify and have a better view regarding this.

Regards,

Aznie

[Ramasabarinathan]

Hello Aznie,

    Thanks for checking on this. Please find the attached model and it is converted from intel distribution under  /intel/openvino/deployment_tools/open_model_zoo/models/public/mask_rcnn_inception_v2_coco

[IntelSupport]

Hi Ramasabarinathan,

As I see, ReLU is not one of the supported layers by dynamic batching features referring to the documentation that I have provided before. This might be the reason for the error.

Regards,

Aznie

[Syamimi_Intel]

Hi Ramasabarinathan,

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

Regards,

Syamimi