@NickM-27 commented on GitHub (Oct 30, 2023):
without hwaccel the CPU usage is going to be much higher. Also worth noting that when hwaccel is used, you can configure frigate to use the GPU to…
@NickM-27 commented on GitHub (Oct 30, 2023):
There's nothing special about it, just follow the docs for using a coral and the docs for setting up hwaccel with nvidia
@FigJam23 commented on GitHub (Oct 30, 2023):
Oh ok so I've already got my Nvidia setup for hwaccel. So this below if I change from
@github-actions[bot] commented on GitHub (Nov 30, 2023):
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.…
@NickM-27 commented on GitHub (Oct 30, 2023):
yes indeed, that would do it
@FigJam23 commented on GitHub (Oct 30, 2023):
Cool thanks, so no model paths needed this is done by the default config for coral I guess
@FigJam23 commented on GitHub (Oct 31, 2023):
Cool made changes I didn't add models left as default seems to be working good, coral is definitely getting close to hot not sure if that's normal.…
@NickM-27 commented on GitHub (Oct 31, 2023):
Coral getting hot is not unexpected, that's why the usb one has a heat sink and the pcie coral supports throttling.
@github-actions[bot] commented on GitHub (Nov 30, 2023):
This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.…
@NickM-27 commented on GitHub (Oct 30, 2023):
The default model in frigate is a model trained on the COCO dataset which as you can see doesn't look anything like…
@nickcmeyer commented on GitHub (Oct 30, 2023):
it is not recommended to run detect on the main stream, it is inefficient and does not provide a benefit in most cases. Also definitely not…
@FigJam23 commented on GitHub (Oct 30, 2023):
Ok cool that sounds like something I would like to do, so is there any info you can point me to so I can set it up like that. Or could you please…
@blakeblackshear commented on GitHub (Oct 30, 2023):
Because the included model is 68% confident that looks like a person. You have lowered your thresholds to 0.6, so you are going to see this…
@NickM-27 commented on GitHub (Oct 30, 2023):
Only one detector type is supported at a time, you can still use the coral for object detection and the GPU for hwaccel (this is what I do). The…
@NickM-27 commented on GitHub (Oct 30, 2023):
it is not recommended to run detect on the main stream, it is inefficient and does not provide a benefit in most cases. Also definitely not…