Porting HRNet to MYRIAD device

[Simardeep]

Hi, 

I recently ported a model with HRNet backbone to openvino ir and was able to inference the model on CPU and GPU. But was unable to inference the model on MYRIAD device. I understand that the openvino documentation displays HRNet as an unsupported model for the MYRIAD device but is there any specific reason for this? Is it due to an unsupported layer or the model size itself?

Regards,

Simardeep Singh Sethi

[Megat_Intel]

Hi Simardeep,

Thank you for reaching out to us.

Yes, you are correct. HRNet models are not supported on MYRIAD and are only supported on CPU and GPU as shown on the Public Pre-Trained Models Device Support page.

For your information, I have verified the hrnet-v2-c1-segmentation model by running Image Segmentation C++ Demo on MYRIAD and received this error:

The "Failed to allocate graph: NC_OUT_OF_MEMORY" error is due to the model size being too big. MYRIAD device has a small memory size, therefore it is unable to load the model.

Other than that, unsupported layers are also a possible reason for the HRNet model not being supported by the MYRIAD plugin. You can refer to Supported Layers for MYRIAD VPU plugin for more information.

Regards,

Megat

[Megat_Intel]

Hi Simardeep,

This thread will no longer be monitored since we have provided a solution. If you need any additional information from Intel, please submit a new question.

Regards,

Megat