Neural Compute Engine: Maximum Utilization and Framerate

[JojoDevel]

I just read here that the Myriad X VPU has 2 Neural Compute Engines. Is that fully utilized with a single neural network (like object detection with Yolov5)?

Can I see the utilization somewhere?

Would it make sense to launch two yolov5 networks and alternately feed a camera image to the first or the second instance? Could that increase the maximum framerate to execute yolov5?

[Erol444]

Hello @JojoDevel , NeuralNetwork node actually has nn.setNumInferenceThreads(threads) and nn.setNumNCEPerInferenceThread(nceNum) (NCE=Neural Compute Engine) functions, so you don't need to run your NN (yolov5) twice, you can just set these 2 functions. Both of these functions have 0 = AUTO value by default. Thanks, Erik