Where did you order from?
I have a search on ebay for the USB coral, and I occasionally get an email from ebay when one is listed. But they are silly prices. The last few USD429, USD499, USD420, USD395.
I did order one from a guy in Japan on ebay at a reasonable price. When it got here, there was some worthless thing like a short string of LEDs in the packet. He refunded me, but that wasnāt really the point.
I ordered it from Seeedstudio.com. Back in november it was listed as backordered but let me put an ordered in so I said screw it and did it.
Looks like right now atleast for the USB model you canāt back order it not sure about the other models
The Coral āDev Board Miniā seems still to be in stock. Does anyone have experience running Frigate on it?
I have seen a post about it. https://github.com/blakeblackshear/frigate/issues/2721#issuecomment-1026466821
1 Like
Checked my order status and delivery pushed out another 6 weeks to 8/31/22ā¦
In your last line you mentioned this: I suspect once I get a coral this will be my way forward given the ease of adding face detection but until I get a coral its to CPU intensive to use as I add more cameras.
By easy to add face detection do you mean by using deepstack with Frigate?
Iām running a pi4 with 4gb ram, currently just running HA, only want to run 3 or 4 cameras and was hoping to get away with just using the Pi (and perhaps a Coral stick when I can get one). Iām not looking to store lots of footage but I do want object detection and facial recognition ideally.
Do you think it would manage all that or am I asking too much? I have other machines I could run stuff on but power is a concern as we have frequent blackouts and the Pi is on backup power where as more powerful (power hungry machines) arenāt.
Thanks
The ease of adding face detection is because of the addon double take which works together very seamlessly with frigate. It automatically pulls events, looks for a person, looks for a face and then tries to match it. All of this can be done with other services but the ācodingā/setup is above my capabilities. With double take you pretty much point it at a camera feed and train faces using the UI.
I think if you got a coral you can use Frigate with a Pi, but double take still uses the CPU so I am not sure how that would work. From my understanding (which could be incorrect) double take only runs detection when it gets an event from frigate so its not a constant loud on the CPU so MAYBE it would work on a Pi just not very fast.
2 Likes
I need some advice, Iām really in integrating NVR etc. now my HA is installed on RAS4, I donāt think I would like to buy now a Coral, BUT i have a server on synology is it possible to run some of the object detection on my synology without moving the HA ?
Frigate does not have to be installed on the same machine as HA.
However I donāt think it can be installed on synology. Not sure though.
Did you find anything?
I have Nvidia Nano sitting unused. Frigate does not support it nowā¦ What I read in the last several days tells me that people are trying to run/compile Frigate for Nvida but I yet to read success stories.
1 Like
Iām afraid I have not found an alternative.
However, Coral AI USB dongles are beginning to be back in stock.
Thanks. If you know where to get USB Coral - please share.
Sightā¦ It would be great if Fregate would be able to use Nano, or there would be any alternative of Fregate but there is none seemsā¦
For instance, it says here https://www.seeedstudio.com/Coral-USB-Accelerator-p-2899.html that itās in stock, although I have not tried it. Iām just going off the links at USB Accelerator | Coral.
Thatās great, been a long time since anyone had stock!
Not relly yetā¦
Considering the very limited supply of Coral USB Accelerator at this stage, only the Frigate NVR bundle kit is available online now. We will open the sale of the Coral USB Accelerator when the supply is back to normal.
Iāve been running Frigate in Docker successfully on a Synology for some time now including Coral functionality. It works excellent on the frigate part. Mind you, running the entire process of person recognition including double-take is quite cpu-hefty and I donāt recommend that sort of workflow usage on a professional level because of its hiccups by system bottlenecks.
Take a look at this repository: GitHub - weltenwort/frigate-synology-dsm7: Dockerfile and docker-compose file to enable google coral USB accelerators in containers on Synology DSM 7
1 Like
Do you mean that for person recognition, in addition to the ML model running on Coral, the additional CPU overhead can become prohibitively high?
I meant: specifically for Synology devices (RS2418+ here). I run the entire workflow of Frigate, Double-Take and Deepstack for object and face recognition and consider this a tad too much for the CPU to handle. Quite possibly this is due to the unavailability of hardware transcoding on many Synology devices like mine.
Frigate itself runs excellent with Coral concerning the movement recognition, but after Frigate detects movement a still image is exported to both Deepstack and Double-Take and those guys will process that image by plain CPU power and thatās when you might hit Synology CPU limitations. Also, Frigate will optionally convert your separate camera streams in a single multi-camera view, which again depends on CPU power.
I my case it all runs on the Synology with three cameras and I need to dial down the camera settings to lower resolutions and lower the stream bitrate and lower the fps to make it all work, and still the occasional movement occurs too quick for the detection to trigger or for the recording mode to initiate in time and with less choppiness.
So I would advise you to test your own workflows and possibly delegate the different tasks to different machines depending on your own test results. Newer machines or different Synology CPUs might yield different results.
I read that there are several developments ongoing on optimizing the whole workflow, ranging from the developer of Frigate considering implementing face recognition directly into Frigate ([FR] Face recognition Ā· Issue #260 Ā· blakeblackshear/frigate Ā· GitHub) to having a Home-Assistant+Double-Take+Coral Addon (GitHub - grinco/HASS-coral-rest-api: Coral REST API for HASS).
1 Like
Hi NL-DUX,
did you ever get the coral TPU rest-API running? I have almost the same setup like you, but i can not start the app properly in hassio due to the lack of configuration description/help fileā¦
Did you get further?
Best Regards,
Alex