YoloV8 Deep Neural Nets on Rock5B SBC using the ARMBIAN OS

In this video, I changed the Operating System from Radxa's official image B39 to Armbian OS which is also approved by Radxa. It is based on Ubuntu x11. I followed this link
to install the Mali driver and then i followed Radxa's links to install the new RKNPU2 driver 0.9.2 together with the latest version of RKNN-toolkit2 v1.6.0.2 and RKNN-Model-Zoo.
Immediately, I could tell that the image processing is much faster on this OS, everything being equal, the #Rock5B is grabbing images from an IP camera connected to a 4G TP-link extender, using YoloV8-rknn to do object recognition and storing the video on a USB stick. Apart from all this, there are multiple applications opened at the same time, as shown in this video. One should also appreciate that even though the internet speed is not phenomenal, the system still manages to achieve quasi-real time processing, with a delay of not more than 0.1s. Again, this is achieved because the NPU is enabled and I installed the Panfrost/fork Wayland GPU driver which should handle part of the image processing. My next attempt is to try to use OpenCL in this application, even though I was told that OpenCL does not work on images , but on matrices. However, an image is made up of a matrix and therefore there must be a way, that I could use OpenCL to enhance the image processing part of this system by using the compute modules of the GPU to do MAC operations in parallel with the NPU and the CPU. I'll keep you posted.