Closed podarok closed 1 year ago
Hi @podarok could you explain a bit more about what you are suggesting here?
Hi @MatthewARM I'm using orange pi 5 which is based on RK3588S CPU for frigate NVR. There is a tensor lite ML detection. With Armnn I was able to accomplish good performance from ARM core and MALI 610 GPU, but NPU - not used. Above I posted a link to library, created by rockchip team that would be great to incorporate in armnn in order to use 6TOPS NPU for ML tasks
Understood. Needs someone to volunteer to add the new backend and maintain it.
For Frigate we succeeded with rknpu2 here https://github.com/blakeblackshear/frigate/pull/5733#issuecomment-1479317334
But this means we should have both armnn and rknpu2 detectors, which is not good long term
I still think rknpu2 is better to have as a part of armnn. But as you said @MatthewARM - depends of human resources at this point
Hi @podarok
I'm going to close this ticket as it's outside of the scope of the ArmNN project.
Best regards, Mike
https://github.com/rockchip-linux/rknpu2 just released
Related issues