I see people using the TPU pie with Frigate for intelligent CCTV. Perhaps you can do a video on how you can capture people littering, throwing litter etc. Or vehicles blocking your entrance etc
@@AdvancedHobbyLab sadly so far I just see it identifying objects, but not really doing anything with that data. It would be great if it could learn when a person litters and then notify you. Record the incident etc. You could also demonstrate the capability of the pie, how many (4k) cameras the pie (+tpu) can handle etc
The USB version of the Coral AI using is more useful for the Raspberry Pi 4 or 5, on the Pi5 you can save the PCIe for use with an NVME drive instead. If you do go the PCIe M.2 route it is worth looking at the new Raspberry PI AI hat which can give you 15TOPS. On another note, the RTX 4090 TOPS rating is 1321 not ~100
It looks like the Raspberry Pi AI hat uses the Hailo-8L processor. I would like to play with one of those if I get the chance. Regarding the RTX 4090, your right! Thanks for the feedback. I was looking at the F32 performance, but the Tensor Core performance is much higher and would have made more sense when comparing to the Google TPU. For completeness, the RTX 4060 supports 242 Tensor Core TOPS.
@@AdvancedHobbyLab I decided to do a quick search in my emails after posting my comment. the AI Hat+ is now available with the Hailo 8 rated at 26TOPS Jeff Geerling did a video attempting to chain PCIe devices, failed due to power issues but still worth looking into.
The Hailo-10H claims to be able to handle LLMs and should have the same M.2 form factor, though I don't think it is released yet. You could also look at one of the Nvidia Jetson dev kits.
Thanks for watching. In my video, I just used a USB webcam so I didn't need the picamera2 library. Fortunately, picamera2 and tensorflow have python interfaces so connecting them shouldn't be a problem. I might do a tutorial in the future about how interface the picamera2 library with the TPU.
For quick questions/comments, the TH-cam comment section is great. For longer discussions, you can message me on facebook. I try to respond as quickly as I can.
@@AdvancedHobbyLab got it, thank you! Maybe then Coral can be directly connected to something like Latte Panda Sigma. It has four M.2 slots, B-Key, E-Key, M-Key, and another M-Key. It is an impressive small computer, just slightly larger than a Raspberry PI. Also has Arduino built-in, check it out.
I see people using the TPU pie with Frigate for intelligent CCTV. Perhaps you can do a video on how you can capture people littering, throwing litter etc. Or vehicles blocking your entrance etc
Thant's a good idea! Thanks for the feedback.
@@AdvancedHobbyLab sadly so far I just see it identifying objects, but not really doing anything with that data. It would be great if it could learn when a person litters and then notify you. Record the incident etc. You could also demonstrate the capability of the pie, how many (4k) cameras the pie (+tpu) can handle etc
The USB version of the Coral AI using is more useful for the Raspberry Pi 4 or 5, on the Pi5 you can save the PCIe for use with an NVME drive instead. If you do go the PCIe M.2 route it is worth looking at the new Raspberry PI AI hat which can give you 15TOPS.
On another note, the RTX 4090 TOPS rating is 1321 not ~100
It looks like the Raspberry Pi AI hat uses the Hailo-8L processor. I would like to play with one of those if I get the chance.
Regarding the RTX 4090, your right! Thanks for the feedback. I was looking at the F32 performance, but the Tensor Core performance is much higher and would have made more sense when comparing to the Google TPU. For completeness, the RTX 4060 supports 242 Tensor Core TOPS.
@@AdvancedHobbyLab I decided to do a quick search in my emails after posting my comment. the AI Hat+ is now available with the Hailo 8 rated at 26TOPS Jeff Geerling did a video attempting to chain PCIe devices, failed due to power issues but still worth looking into.
Damn i was hoping to use the coral TPU to set up local LLM's maybe with crewAI or similar 😕 are there maybe alternatives?
The Hailo-10H claims to be able to handle LLMs and should have the same M.2 form factor, though I don't think it is released yet. You could also look at one of the Nvidia Jetson dev kits.
@@AdvancedHobbyLab I'll have a look, thanks for the heads up
@@AdvancedHobbyLab TY for this answer, I spent far too long looking for videos of using the hailo with a llm before seeing your comment here!
thanks for this tutorial . I have questions about how you operation the camera and how install picamera2 with edge ?
Thanks for watching. In my video, I just used a USB webcam so I didn't need the picamera2 library. Fortunately, picamera2 and tensorflow have python interfaces so connecting them shouldn't be a problem. I might do a tutorial in the future about how interface the picamera2 library with the TPU.
Hey man I have a question about kinematics you designed for a spot micro project what is the best way to contact you?
For quick questions/comments, the TH-cam comment section is great. For longer discussions, you can message me on facebook. I try to respond as quickly as I can.
I thought Coral was discontinued. Is this a newer model? Are they back in the game?
The Coral TPU has not been discontinued, however, not too long ago, they had supply chain issues and were really hard to find.
@@AdvancedHobbyLab Thank you. It seems like there are other M.2 versions, would one of them fit Pi5 directly?
As far as I know, there is no way to attach a TPU to a Pi without an M.2 adapter board.
@@AdvancedHobbyLab got it, thank you! Maybe then Coral can be directly connected to something like Latte Panda Sigma. It has four M.2 slots, B-Key, E-Key, M-Key, and another M-Key. It is an impressive small computer, just slightly larger than a Raspberry PI. Also has Arduino built-in, check it out.
Wow, that's cool! Coral AI makes a dual edge TPU with an E-Key that looks like it should work with the Latte Panda Sigma. Thanks for sharing!