YOLOv11 Model Naming Support

[Lunar-Academy]

From my testing, creating a YOLOv11 model and quantizing/converting to rknn works with PhotonVision's object detection. However, the model must be named as a YOLOv8 model. From looking at the RKNN toolkit-2 examples (RKNN Model Zoo), the way of running the models seem identical. I wasn't able to find any differences in the files (YOLOv8 and YOLOv11) used for processing the models. Because of this, PhotonVision's backend for YOLOv8 detection works for YOLOv11.

Model I have been testing with (Github issues doesn't let me upload the file directly, so it's a link to my team's code.)

[samfreund]

This is currently in progress, I'm working on integrating this into the JNI right now.

[Lunar-Academy]

While you're doing that, do you think you would be able to modify the npu core selection? From what I can tell, PV currently relies on autoselecting the npu cores.

RKNN Object Detector file

If the -1 on core selection parameter of the create method were be changed to 210, theoretically PV would be able to effectively utilize all npu cores.

[samfreund]

I believe that's currently addressed in another PR. Not quite sure of the number. #1744

[Lunar-Academy]

Ah, got it