GPU support for object detection using TensorFlow Lite in my own custom image on the imx8dxp chip.

[Yu-ki]

Hi, Team

I am currently developing on my own custom image on imx8dxp. When using TensorFlow Lite for object detection, I am running it only on the CPU and now I want to run it on the GPU. I found in the i.MX Machine Learning User's Guide that I can use the vx-delegate to enable TensorFlow Lite with GPU support. Can I enable GPU support by building the library from the following repository and setting the path?

https://github.com/nxp-imx/tflite-vx-delegate-imx.git

best regards,

[brian14]

Hi @Yu-ki, 

Thank you for contacting NXP Support.

Yes, you can try building the VX Delegate, however, the recommendation is to use The Yocto Project recipe specifically the .bb file for TIM VX.
meta-imx/meta-ml/recipes-libraries/tim-vx/tim-vx_1.1.42.bb at mickledore-6.1.22-2.0.0 · nxp-imx/meta...

Base your building process on that .bb file.
Here is the TIM VX repository for more information:

GitHub - nxp-imx/tim-vx-imx at lf-6.1.55_2.2.0

Have a great day!