Is it possible to use the jetson nano with photon vision for note object detection? Is this possible with a coral or on the cpu or is it better to just get an orange pi?
1 Like
Photonvision object detection uses the rockchip npu, which is only available on orange pi out of those devices. Coral only runs tflite models, but photonvision uses the npu with rknn models.
1 Like
Our team uses a Jetson Nano for object detection. We use this Ubuntu image which comes with Tensorflow and Pytorch preinstalled. From there we run the pretrained YOLOv8 Dataset Colab model.
1 Like
It’s worth noting here that the Jetson nano is EOL from nvidia and is not supported by them anymore, and wpilib will not run natively due to GCC version mismatches. You can to use docker to hack around this by running a VM with a newer glibc, but it’s also a lot to deal with unless you’ve used Docker before
We just found out that the image for the OPi 5 from PV runs Object Detection on the NPU this past week. And I must say, we are quite impredded. It just works! Even without tuning, we are getting 30fps with approx 15ms lag.
To answer your question, based on what @thatmattguy pointed out, and on our experience, it is VERY WORTH getting an OPi 5 if possible!
Ive got a working jetson with photon vision do I need to use the pre installed image or is it too difficult to do by hand with the os ive got?
Also, are you processing april tags with the jetson along with the object detection?
Thanks! our team has 2 jetsons so getting a solution working on them would be best but if it doesn’t work i’ll bring it up to get an orange pi.
Oh, I am sorry, I misunderstood, if you have a working jetson with photonvision is probably best to use that over the image I provided.
We also have two jetsons and use one for apriltags and one for object detction. It works very well and we were able to get 4 almost hd (compressed to 88% of the original size) streams into our apriltag system.