Possible to use substream for detection but mainstream for live view?

[jonadis]

I think I've got things setup so detection uses the lower resolution substream and the recordings use the full resolution mainstream, but it looks like the live view and birdseye view use the lower resolution substream. Is it possible to leave the detector using the substream but have the live view use the mainstream?

  frontporch:
    ffmpeg:
      inputs:
        - path: rtsp:https://admin:[email protected]:554/cam/realmonitor?channel=1&subtype=0
          roles:
            - record
            - rtmp
        - path: rtsp:https://admin:[email protected]:554/cam/realmonitor?channel=1&subtype=1
          roles:
            - detect
    detect:
      width: 640
      height: 480
      fps: 5

[blakeblackshear]

It's not. Birdseye and live view use jsmpeg for low latency viewing which requires decoding the stream.

[thunderwolf66102]

Could it be a configurable option for live view to use a different feed via roles - via a liveview role? I would imagine decoding/encoding for birdseye is most of the work/cpu load and it may be possible just to copy the stream directly from the camera for the live view? It would be nice to stay within the frigate interface and not have to open another program or log into a different web page to get a max resolution view.

Or a toggle between inputs within live view?

[MrColumbo]

I wanted to accomplish the same and even thought the rtmp role is used for that.
I am confused about the answer ...basically I do not understand the reason why live view is always showing the stream with the detect role .....why does it not be that stream?

[NickM-27]

I wanted to accomplish the same and even thought the rtmp role is used for that. I am confused about the answer ...basically I do not understand the reason why live view is always showing the stream with the detect role .....why does it not be that stream?

RTMP is just used for sharing the stream to other clients (like HomeAssistant). The frigate live view uses the detect role as it already needs to be decoded for object detection. Doing the above would require decoding both streams from each camera constantly which would result in GREATLY increased CPU usage

[elektrinis]

In my case I have a 4K main stream and a 640x360 substream, and there are no other resolutions to pick.
Motion detection works fine, however smaller objects like cats and dogs just don't work. Even humans and cars are being detected with poor accuracy, like, a manhole is identified as a person. Example below.

Is it possible to use the substream for detection (light on CPU), and then use the main stream, crop the frame to size without analyzing it and passing that for object recognition? I know 4K is CPU intensive, but if it's only on motion, I don't see the harm.

[Caddyman68]

I have a similar situation. Camera is set to record at main resolution (4k) and detect at 640x480. I can't view the recordings on the app unless I record at the 640x480 which would be fine for a quick look on my phone, but if I need to see minute detail such as a license plate on my PC screen it will be the 640x480. I may try to use 2 camera instances for each camera where the app cam records at 640x480 and the "record cam" records at full resolution. Sounds like it will be resource (CPU) intensive and maybe not even work though.

[chaosye0]

I have same problem, how to show the recording stream in web ui rather than the stream for detect?

[shanekuz]

I have the same problem would like live view and home assistant card to show full stream not detect stream