Sorry maybe a daft question but can I plug in a google coral usb and configure deepstack to utilize the analyzing power of coral instead of the cpu? Been reading around for a while and can't find a definitive answer as it all relates to frigate.
haha, Funny I was just playing with frigate earlier today. Just found out there is such thing ar frigate and coral a day or two ago. I was wondering the same thing if somehow we can use Coral for the Processing of Deepstack anyhow I think frigate so far seems to be cool but a bit complex for me to setup..
I guess will get there, I have so far manage to run one of the Dahua cams on frigate under HA but no object detection yet or coral stuff or How / where to dump the recording as this is all new to me.
haha, Funny I was just playing with frigate earlier today. Just found out there is such thing ar frigate and coral a day or two ago. I was wondering the same thing if somehow we can use Coral for the Processing of Deepstack anyhow I think frigate so far seems to be cool but a bit complex for me to setup..
I guess will get there, I have so far manage to run one of the Dahua cams on frigate under HA but no object detection yet or coral stuff or How / where to dump the recording as this is all new to me.
I'm very happy with BI and deepstack but see impressive analysing speeds of this coral accelerator and need to know if I can get deepstack to use it.
My gpu is amd so not compatible with deepstack so uses my main cpu power.
Yeah, I think I was reading somewhere Google Coral USB does has support for Windows 10 platform .. and can do something called MobileNet v2 at 400 FPS of processing, Again dont have much idea about it as I stumbled upon this couple of days ago but this is interesting for sure. I was also thinking if somehow that coral can be used for Open ALPR processing...
Hope someone else may have better idea on this and may be can guide us to the right direction if its even possible not sure if it is at this point or no..
I had a small atom intel pc spare (emmc memory) 4gb - threw debian on it and installed deepstack. Sent the processing through this and it was soo slow. 2-3.5 seconds per snapshot!!
At least on my ryzen 5 (which I know is not optimized for deepstack), takes about 350-700ms per photos. normally about 500ms.
I need to get my intel dell tower i7 back up and running as that was faster about 2-300ms per HD snapshot. Out of action right now.
To answer the OP directly. Deepstack has no support for Coral TPU at this time.
There was a discussion opened on DeepStack forum that ended in a "not likely" Is coral support likely? - DeepStack Forum
That said, wouldn't hurt to look deeper on DeepStack github and see if anyone had opened a request for it.
I find Deepstack does a pretty good job on its own using CPU or GPU processing, but I don't have a high volume to process. With a high power CPU / GPU DeepStack can do quite high volumes I believe, but not on low power devices. If you have a high volume environment, unless you specifically need DeepStack, it may be worth looking at other options that use Coral already.
