Need some guildence

[keithsplace1]

Here is set up Frigate running on Zima OS and also same set up in UmbrelOS.
Zima has a nvidia GPU (Coral TPM) on the way. The other has no GPU, twice the RAM but could use Coral.
Here the config working with 6 Cameras only two same brand. 2-KKmoon PTZ onvif, 1-Hik knockoff Doorbell, 1-Hikv Dome,
1-Yucheng motion tracking PTZ onvif. Also three are H256 and two are H264, so I'm assuming I don't need trans coding?

Having trouble getting all the streams but I'm finding them by trial and error, more errors.
but I actually have all working just not all the correct streams so far.
What I want is to detect low res streams, record motion or object detects only, all these cams are on a single network connection to an NVR but it only records 24/7. But does stream to my phone from anywhere nicely, just trying to reduce network traffic.

Here is what I have if you could tell me what the settings should be, would be awesome.

Here is what I need to know:
How to write the High res stream to record lines?
And how to write the GPU to do trans-coding and the Coral to do detection/object stuff, when it arrives.
Doorbell has 1 stream so not worried about that one.
Short: how to set GPU
How to set Coral
How to record High res motion detection.

mqtt:
enabled: false

cameras:
Office: # <------ Name the camera
ffmpeg:
inputs:
- path: rtsp://192.168.1.222:554/11 # <----- The stream you want to use for detection
roles:
- detect
- record
detect:
enabled: false # <---- disable detection until you have a working camera feed
width: 1920 # <---- update for your camera's resolution
height: 1080 # <---- update for your camera's resolution
detect:
enabled: false

Drive: # <------ Name the camera
ffmpeg:
inputs:
- path: admin:xxxx@rtsp://192.168.1.219:554/11 # <----- The stream you want to use for detection
roles:
- detect
- record
detect:
enabled: false # <---- disable detection until you have a working camera feed
width: 2560 # <---- update for your camera's resolution
height: 1440 # <---- update for your camera's resolution
detect:
enabled: false

Doorbell: # <------ Name the camera
ffmpeg:
  inputs:
    - path: rtsp://192.168.1.122:554/1Stream/Channels/1/Main # <----- The stream you want to use for detection
      roles:
        - detect
        -  record
detect:
  enabled: false # <---- disable detection until you have a working camera feed
  width: 2560 # <---- update for your camera's resolution
  height: 1440 # <---- update for your camera's resolution

detect:
enabled: false
version: 0.16-0

[inkeep-ai-assistant[bot]]

Based on your setup, here's guidance for configuring Frigate with separate detect and record streams, GPU acceleration, and Coral TPU.

Correct Configuration Structure

Your current config has several issues. Each camera should be properly indented under cameras:, and detect should only appear once globally or at the camera level(1).

Setting Up Dual Streams (Low-res Detect + High-res Record)

For cameras with multiple streams, configure separate paths for detect and record roles(2):

cameras:
  Office:
    ffmpeg:
      inputs:
        - path: rtsp://192.168.1.222:554/11  # Low-res stream for detection
          roles:
            - detect
        - path: rtsp://192.168.1.222:554/01  # High-res stream for recording
          roles:
            - record
    detect:
      enabled: true
      width: 640   # Match your low-res stream
      height: 360
      fps: 5

(3)(2)

Configuring NVIDIA GPU Hardware Acceleration

For NVIDIA GPU transcoding, add this to your global ffmpeg section(4):

ffmpeg:
  hwaccel_args: preset-nvidia

(4)

You'll also need to pass the GPU to the Docker container. Add to your docker-compose.yml:

services:
  frigate:
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

(3)

Configuring Coral TPU

When your Coral arrives, add this detector configuration(1):

detectors:
  coral:
    type: edgetpu
    device: usb

(1)

And pass the USB Coral to Docker:

services:
  frigate:
    devices:
      - /dev/bus/usb:/dev/bus/usb

(3)

Recording Motion/Objects Only

Configure recording to save only when motion is detected(1):

record:
  enabled: true
  retain:
    days: 7
    mode: motion

(1)

H.264 vs H.265 Transcoding

Since you have both H.264 and H.265 cameras, you don't need transcoding for playback if your clients support both codecs. However, if you need Apple compatibility for H.265 streams, enable this(1):

ffmpeg:
  apple_compatibility: true

(1)

Complete Example Configuration

mqtt:
  enabled: false

ffmpeg:
  hwaccel_args: preset-nvidia

detectors:
  coral:
    type: edgetpu
    device: usb

record:
  enabled: true
  retain:
    days: 7
    mode: motion

cameras:
  Office:
    ffmpeg:
      inputs:
        - path: rtsp://192.168.1.222:554/11
          roles:
            - detect
        - path: rtsp://192.168.1.222:554/01
          roles:
            - record
    detect:
      enabled: true
      width: 640
      height: 360
      fps: 5

  Drive:
    ffmpeg:
      inputs:
        - path: rtsp://admin:xxxx@192.168.1.219:554/11
          roles:
            - detect
        - path: rtsp://admin:xxxx@192.168.1.219:554/01
          roles:
            - record
    detect:
      enabled: true
      width: 640
      height: 360
      fps: 5

  Doorbell:
    ffmpeg:
      inputs:
        - path: rtsp://192.168.1.122:554/Stream/Channels/1/Sub
          roles:
            - detect
        - path: rtsp://192.168.1.122:554/Stream/Channels/1/Main
          roles:
            - record
    detect:
      enabled: true

(3)(1)(2)

Note: You'll need to find the correct RTSP paths for your camera's substreams (typically lower resolution). Common patterns are /11 for main and /12 for sub, or /main and /sub(2).

📚 Sources:

1. Full Reference Config
2. Camera Configuration
3. Getting Started
4. FFmpeg presets

---

Have another question? Just tag @inkeep.