Using Go2RTC to transcode to h.264 but recordings are still in h.265

[Foritus]

Hello friends!
My camera only provides H.265 feeds and they don't play in Firefox. So I followed the instructions to use Go2RTC to transcode the feeds into H.264 and the live view now works perfectly. However the recordings are somehow still being written in the original H.265 🤔

Is there a way for me to save the transcoded feed instead? Otherwise viewing history doesn't work in non-hevc-capable browsers :(

Config:

# Reference config: https://docs.frigate.video/configuration/reference/
version: 0.15-1

mqtt:
  enabled: false

ffmpeg:
  hwaccel_args: preset-vaapi
  output_args:
    record: preset-record-generic-audio-aac

detect:
  enabled: true

detectors:
  coral:
    type: edgetpu
    device: usb

database:
  # The path to store the SQLite DB (default: /config/frigate.db)
  path: /db/frigate.db

record:
  enabled: true
  # Sync recordings with disk on startup and once a day (default: false)
  sync_recordings: true

  retain:
    days: 3
    mode: all

  alerts:
    retain:
      days: 30
    pre_capture: 5
    post_capture: 5
  detections:
    retain:
      days: 30
    pre_capture: 5
    post_capture: 5

go2rtc:
  streams:
    rear-garden-high-res:
      - rtsp://creds:creds@hostname/h265Preview_01_main
      - "ffmpeg:rear-garden-high-res#video=h264#audio=copy#hardware"
    rear-garden-low-res:
      - rtsp://creds:creds@hostname/h265Preview_01_sub
      - "ffmpeg:rear-garden-low-res#video=h264#audio=copy#hardware"

cameras:
  reargarden:
    enabled: true
    ffmpeg:
      # Use QSV for H.264 here as we're consuming the post-Go2RTC feed here, I think?
      hwaccel_args: preset-intel-qsv-h264
      inputs:
        - path: rtsp://127.0.0.1:8554/rear-garden-high-res
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/rear-garden-low-res
          input_args: preset-rtsp-restream
          roles:
            - detect
    motion:
      # Enables dynamic contrast improvement. This should help improve night detections at the cost of making motion detection more sensitive
      # for daytime.
      improve_contrast: false
      # The threshold passed to cv2.threshold to determine if a pixel is different enough to be counted as motion. (default: shown below)
      # Increasing this value will make motion detection less sensitive and decreasing it will make motion detection more sensitive.
      # The value should be between 1 and 255.
      threshold: 30
      # Minimum size in pixels in the resized motion image that counts as motion (default: 10)
      # Increasing this value will prevent smaller areas of motion from being detected. Decreasing will
      # make motion detection more sensitive to smaller moving objects.
      # As a rule of thumb:
      #  - 10 - high sensitivity
      #  - 30 - medium sensitivity
      #  - 50 - low sensitivity
      contour_area: 30
    live:
      stream_name: rear-garden-low-res
    ui:
      # Optional: Adjust sort order of cameras in the UI. Larger numbers come later (default: 0)
      # By default the cameras are sorted alphabetically.
      order: 1
    birdseye:
      # Optional: Adjust sort order of cameras in the Birdseye view. Larger numbers come later (default: 0)
      # By default the cameras are sorted alphabetically.
      order: 1

[hawkeye217]

You should provide a single stream only from go2rtc, eg:

go2rtc:
  streams:
    rear-garden-high-res:
      - ffmpeg:rtsp://creds:creds@hostname/h265Preview_01_main#video=h264#audio=copy#hardware

This creates a single transcoded stream from go2rtc (rather than the original and transcoded one) that gets fed into Frigate.

[hawkeye217]

What CPU do you have? Transcoding is always going to cause a CPU hit of some kind, even with acceleration.

[Foritus]

Wait, I suspect this was due to a typo. I left a trailing quote mark on the end of the Go2RTC source line so it ended with #hardware"

[Foritus]

Yes that fixed it. CPU usage is normal again. You can tell I need a nap today 😄

All working, thank you so much!

[hawkeye217]

Great! Glad it's working. Happy to help!

[Foritus]

So this exposed a bug in Intel-VAAPI with the high resolution feed from my camera, where it would error out with a bizarre decode error:

intel vaapi Error submitting packet to decoder: 
Input/output error Could not find ref with POC 1
Failed to end picture decode issue hardware accelerator failed to decode picture

This doesn't happen when using QSV on this system (Celeron N5105, which has a gen 11 / ice lake iGPU), but Go2RTC only supports VAAPI when invoking ffmpeg (I checked the source code https://github.com/AlexxIT/go2rtc/blob/master/internal/ffmpeg/hardware/hardware.go#L14 ).

So instead I've swapped back to using the "native" Go2RTC backend and have instead written a cron job that scans through all the recordings and re-encodes them into h.264 instead 😅 It'll do for now (mostly leaving this comment for debugging-SEO purposes 😄 )