We are working on optimizing our model performance on vpu, however, the complex dataflow fusion make us hard to interpret the output performance counter results. I see there is a DAG viewer based on netron available.
I am wondering is that possible to export the graph optimized by the VPU plugin, and is the netron viewer that can decode the graph publicly available.
What type of graph format are you expecting to get?
We currently do not have the feature to get the device specific conversion of the IR model.
The IR model is the file that is used with OpenVINO as this format is optimized to be translated to multiple devices on the platform.
Currently you can observe the resultant execution on the VPU using the DL workbench
Thank you for your question. If you need any additional information from Intel, please submit a new question as Intel is no longer monitoring this thread.