My team is looking into switching from limelight os to photonvision, how can we convert our old TFLITE models to RKNN, and do we still need the Google coral plugged in?
RKNN is not for limelight.
Yes, that is why i need to convert our tflite models we use for our limelight, to rknn models for photonvision
RKNN is the file type for use with ROCK CHIP processors (Orange Pi5 with built in hardware accelerator for Machine Learning models).
The limelight is not based on a ROCKCHIP main chip. Simply converting your TFLITE model to RKNN, loading PHOTONVISION to your LimeLight and trying to install the RKNN model will not function well (if at all), and will not run on the Google Coral.
If you want to use LimeLight hardware with machine learning you need to use a Google Coral and a TFLITE or Google Coral comparable ML Model.
Even when running photonvision? I’m just trying to follow the instructions they give
Only ways for AI/ML/Object detection:
Limelight with limelight software and a coral
Orange pi or other rk3588 device with PV
1 Like
It’s not a software issue, it’s hardware. As said above, the limelight physically cannot run a rknn model well. Models are compiled to run on a specific device, and a rknn model only can run well on a rock chip device like the Opi5.
the limelight cant run an ML model on photon vision anyways. i tried to do it a while ago and i coudnt see any way to do it, the only way to do so with photon vision is with an Orange Pi 5. but it runs pretty well on it.
1 Like