tflite-vx-delegate-imx on imx8x series and perform inference using GPU?

[Yu-ki]

I'm not very familiar with embedded system development, so maybe I'm not asking the question well.

Can I run sensorflow-lite inference (e.g. object detection) on GPU even if I build libvx_delegate.so using tflite-vx-delegate-imx on host PC and transfer it to imx8qxp chip via SCP? imx8qxp's distribution is using WLX21.

[Bio_TICFSL]

Hello,

Yes you can, you can check the machine learning guide on how to do it. but you migth check the GPU in the MX8QX since it does not have NPU the VX_delegate won't work.

https://www.nxp.com/docs/en/user-guide/IMX-MACHINE-LEARNING-UG.pdf

Regards

[Yu-ki]

Hello, Bio_TICFSL

Thank you for reply,

So does this mean that the GPU on the imx8qxp chip cannot be run using tensorflow-lite?
I thought the GPU can run the GPU (GC7000 Lite) using VX_Delegate, TIM-VX and OVXLIB, 
looking at Figure3 in https://www.nxp.com/docs/en/user-guide/IMX-MACHINE-LEARNING-UG.pdf,