All the time. I'm talking about the frigate "home" page with all the cams "live". I can run VLC with the exact same url frigate uses alongside my browser at any time and see this. Now, it could be that ALL cams have this latency but I can't easily do the same test on any other cam because they are all outside.
Unfortunately I am by far a Frigate export. Only thinking out loud here in this discussion, while struggling to get it up and running. And to allow others to follow my painfull journey, so they can learn from it.
There is a lot of stuff that can cause high latency. It depends on a LOT of things: usage of go2rtc or not, high cpu usage on raspberry, using hardware acceleration or not, and so on and so on...
I think it is best to open a discussion on Reddit or in the Frigate Github repo, and share your config (without private stuff like hostnames, usernames, passwords, ... of course).
I have no idea anymore what I am doing wrong.
So I have opened a new discussion opened in the Frigate repo.
Hopefully they see which mistake I made...
Ok I finally got the permission to create a pull request. But it is a draft PR, to allow them to review my changes. Because there is a concern that I break the HomeAssistant integration, which I completely understand. Hopefully they can improve my implementation and it doesn't get rejected. Because it took me quite a lot of my limited free time to get it running...
Got some great feedback from Blake. I have implemented all his remarks, and it still seems to be working fine. One small step for me, but a giant leap for Frigate via Tailscale
Blake - who lives in the US - again did some extra suggestions tonight. I have implemented those quickly, and still everything seems to be working fine. It looks like we are near to a working feature.
Haleluja...
Got goodback from the Frigate team again, and - allthough they were very kind - they must have thought by themselves: what a noob ...
I made 2 noob mistakes:
-
The "Explore" tabsheets shows all objects that have been tracked, while the Review tabsheets shows all the review items. Since the persons and cars where detected (outside of my driveway zone) and tracked, they appeared in this tabsheet.
However they didn't appear in the review tabsheet because they didn't intersect with my driveway zone, so NO review item has been created.
-
I got a snapshot for those items, because I didn't no that I had to tell Frigate only to create a snapshopt image when the tracked object intersects with my zone.
I have updated my diagram a bit to make this a bit more clear:
I have updated my private "cheat sheet" a bit more in detail:
Looks quite a lot like a flow with the video related nodes that I have been building in the past...
WRONG! When a replacement card behaved exactly the same way I reported it to the devs, who first thought it was a solved python issue, but after applying their patch the problem just moved to a different line of code. When I reported that they replied that "The Coral getting "stuck" is the bigger concern here, actually."
It appears that plugging the Coral into a usb3 port, whether usbc or usba, doesn't work currently while I have NO problems as long as it's plugged into a usb2 port.
And this is not true with the Coral plugged into a usb2 port. With that the latency in frigate is less than a second longer than in VLC.
Finally my pull request my pull request has been approved by one of their developres. I hope it will be merged soon in their "dev" branch. So I 'expect' it to be released only in their 0.16.0 version.
I have added both ways to their documentation, because the http header mechanism was also not documented yet:
They mention Tailscale all over the place in their documentation, so I have done that also here.
That was more free time consuming as I had hoped. Now finally back time for some other stuff...
The story continues...
Someone reported that I had introduced a breaking change in their dev branch
You know how it works: you do a lot of tests to cover as much scenario's as possible. And then near the end somebody asks you to move a little code snippet to another location. It looks harmless and then you forget to retest one of the scenario's. Oeps...
Anyway I think my last fix now covers the base path setting in all different scenarios. Once the guy has tested it again, I will create a new pull request. Hopefully I get time then for something more fun stuff...
BTW if you follow the commit history of the dev branch on the frigate repo, it is quite mind blowing how much these 2 guys deliver in their free time for the next major release. There are 3 people involved, but it looks like that Blake itself is focussed mainly on the (paid) Frigate+ stuff.
Looking forward to have the next 0.16 release running, to be able to play with all those nice new features. But I will have to wait for a couple of months, before it is released.
I did look at it a bit and noticed more than a few comments about a release 1.0. I didn't see a timeline anywhere though.
I signed up for a free Frigate+ account and I've been submitting snapshots for about a week now. It appears that my local model is benefiting from these submissions but I don't see how that happens so maybe it's just wishful thinking.
The dev branch in their repo is used for the next major release, in this case the 0.16.0 release. Currently it is in alpha status. And afterwards there will be a few beta releases, and I expect a final release somewhere during the summer. I find it quite interesting to see what fun stuff is arriving, although of course I read through the commit history with my developer goggles on...
I have tested their own proposal for the webmanifest fix (see pull request) and it is now all fixed. Hopefully I can now finally continue with some more fun stuff...
Famous last words with a touch of Deja Vu!