I tried opening the streams on iPhone, iPad and MacBook in safari. The video ist way behind for me. When I check the Reolink app I am about 2-3 seconds behind. Everything is running fluid and smooth.
When I check the stream inside frigate add-on it is also around 2-3 seconds behind but stuttering sometimes. When I include the video from frigate in my dashboard ist is more then 10-15 seconds behind.
Not sure if my machine is a problem or something else. Or what I can do to optimize performance.
It needs also time to switch from low res to high res video. I detect person and car on camera, that works well. But also need around 10-15 seconds.to detect people. I use coral usb stick here for frigate.
Could a separate machine for object/face detection and handle the video streams help?
Also loading records takes forever or is crashing the app. Maybe a problem with the custom frigate card?
I know this is all anecdotal, but I have 14 Hikvision IP cameras (10 of which are connected to Frigate for object detection and recording via RTSP from 3 freestanding NVRs) plus 14 Ring cameras (via Ring-MQTT) all integrated into my Home Assistant Dashboard with Lovelace webrtc-camera (IP cams) and picture glance cards (Ring cams) and also use Double Take for person recognition, and everything seems to run pretty smoothly on my HP G4 1L mini desktop Intel i5-8500T with 32 GB RAM and a USB Google Coral device, which is also running HA OS. Rarely do I go above ~25% CPU, 25% RAM use, or 20 Watts power consumption, and it is typically much lower.
I find the client device is often the rate-limiting step. An iPhone 15 Pro and M1 Macbook Pro can just barely handle viewing my setup. 12th gen Intel and Ryzen 5 notebooks work okay but is browser dependent with Edge working best on Windows and Firefox on Ubuntu not at all (even on the same hardware). The dashboard view with camera displays is typically not the problem, either, but rather other dashboard views incorporating graphs and 3D objects seem to be the bottlenecks.
About double take, what face detection software do you use? I have Codeproject.ai because I can only use cpu detection and I am absolutely not happy with it. Detecting face that are not the right person for example. I barely have it detect the right face. When it has an unknown face it uses just any of the existing that could match.
And about performance, yeah I think the HA IOS app is not working well here in compare to the device apps for cameras itself.
I have problems with deepstack crashing very often. And the original website seems to be down. Can you share the link to the repo where you added the addon from.
