How to accelerate yolo v4-tiny?


I am running ubuntu 22.04 downloaded via OOWOW and installed ros2-humble to run darknet_ros from here.

I was able to build the package without CUDA and CUDNN support and can launch yolov4-tiny package however was too slow and eventually crash after a few seconds.

It seems like there is no ai acceleration working.
Is there a way to utilize GPU for running YOLO?
Any suggestions would be much appreciated.

if YOLO requires CUDA your doomed GPU’s are all different by brand and CUDA = Nvidia.
see Nvidia terga (that is Nvideas arm gpu i think)

or look into Yolo GPU compatibility list. in theory any gpu can number crunch but it might require a world class coder to make it compatible with someone else’s code.
and that ain’t me.

you will need software that can do what you need that is compatible with the VIM4 GPU
I’d start with vim4 npu docs i looked for it but see what i thought you would need (at a glance)
but Khadas Docs are amazingly in depth it should be there.

good luck and keep me posted I have an edge 2 and will be working on similar projects.

@lospilot99 Thank you for your comment.
I am currently looking at this document

I followed the instruction and after manually installing opencv 3.4.1, finally can build nnpack version of darknet sucessfully.
Though I still get error when running ./darknet detect test with weights.
I will post my progress here once I get some meaningful outcome.