Frigate, Compreface, Double Take, Poor Quality

Hello everyone.
My setup is an HP EliteDesk 800 G4 DM Core i5-8500T 2.1 - SSD 480 GB - 32GB, with Proxmox installed, HA in VM and Double Take as addon, Frigate in LXC container, and Compreface in LXC container.
I have 6 cores attributed to HA VM, 6 cores and 4GB RAM to Frigate, and 6 cores and 8Gb RAM to Compreface.
Also have Coral USB used by Frigate.
I have a Dahua NVR with 8 cameras, all of them with H.265 video compression, 7 of them with 2560x1440 primary stream (can have better resolution) and 704x576 secondary stream (is the best resolution I can have). The 8th camera has 1280x720 primary stream and 1920x1080 secondary stream.
Frigate is 0.12.0-27A31E7 version.
My Frigate configuration.yml is this:

mqtt:
  host: 192.168.XX.XX
  port: 1883
  user: xxxx
  password: XXXXXX

detectors:
  coral:
    type: edgetpu
    device: usb

go2rtc:
  streams:
    Varanda_da_Frente:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=1&subtype=0
    Varanda_da_Frente_Sub:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/4/cam/realmonitor?channel=1&subtype=1
    Varanda_da_Entrada:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=2&subtype=0
    Varanda_da_Entrada_Sub:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=2&subtype=1
    Lateral_Quartos:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=3&subtype=0
    Lateral_Quartos_Sub:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=3&subtype=1
    Traseira:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=4&subtype=0
    Traseira_Sub:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=4&subtype=1
    Lateral_WC:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=5&subtype=0
    Lateral_WC_Sub:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=5&subtype=1
    Lateral_Rampa:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=6&subtype=0
    Lateral_Rampa_Sub:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=6&subtype=1
    Frente_Portao:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=7&subtype=0
    Frente_Portao_Sub:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=7&subtype=1
    Campainha:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=8&subtype=1
    Campainha_Sub:
      - ffmpeg:rtsp://yyy:[email protected]:xxx/cam/realmonitor?channel=8&subtype=0

birdseye:
  enabled: True
  quality: 8
  mode: objects

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

logger:
  # Optional: Default log verbosity (default: shown below)
  default: info
  # Optional: Component specific logger overrides
  logs:
    frigate.app: debug
 
ffmpeg:
  hwaccel_args: 
    - -hwaccel
    - vaapi
  # Optional: global ffmpeg args (default: shown below)
  global_args: -hide_banner -loglevel warning -threads 1
  # Optional: global hwaccel args (default: shown below)
  # Optional: global input args (default: shown below)
  input_args: -avoid_negative_ts make_zero -fflags +genpts+discardcorrupt -rtsp_transport tcp -stimeout 5000000 -use_wallclock_as_timestamps 1
  # Optional: global output args
  output_args:
    # Optional: output args for detect streams (default: shown below)
    detect: -threads 1 -f rawvideo -pix_fmt yuv420p
    # Optional: output args for record streams (default: shown below)
    record: preset-record-generic

cameras:
  Varanda_da_Frente:
    ffmpeg:    
      inputs:
        - path: rtsp://127.0.0.1:8554/Varanda_da_Frente
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Varanda_da_Frente_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: Varanda_da_Frente_Sub
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25
      stationary:
        interval: 0
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    objects:
      track:
        - person
        - cat
        - dog
        - knife
        - cell phone
        - mouse
      mask: 0,0,1000,0,1000,200,0,200
      filters:
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.7
          mask: 0,0,1000,0,1000,200,0,200
    motion:
      threshold: 25
      contour_area: 30
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 50
      mask: 0,900,1080,900,1440,1920,0,1920
      improve_contrast: False
    best_image_timeout: 60
    # Optional: 24/7 recording configuration
    record:
      # Optional: Enable recording (default: global setting)
      enabled: True
      # Optional: Number of days to retain (default: global setting)
      retain:
        days: 3
        mode: motion
      events:
        retain:
          default: 7
          mode: active_objects
          objects:
            person: 50
    # Optional: RTMP re-stream configuration
    rtmp:
      # Required: Enable the live stream (default: True)
      enabled: False
    # Optional: Configuration for the jpg snapshots written to the clips directory for each event
    snapshots:
      # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
      # This value can be set via MQTT and will be updated in startup based on retained value
      enabled: False
      # Optional: print a timestamp on the snapshots (default: shown below)
      timestamp: False
      # Optional: draw bounding box on the snapshots (default: shown below)
      bounding_box: False
      # Optional: crop the snapshot (default: shown below)
      crop: False
      # Optional: height to resize the snapshot to (default: original size)
      height: 175
      # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
      required_zones: []
      # Optional: Camera override for retention settings (default: global values)
      retain:
        # Required: Default retention days (default: shown below)
        default: 10
        # Optional: Per object retention days
        objects:
          person: 15
    mqtt:
      enabled: True
      timestamp: True
      bounding_box: True
      crop: True
      height: 500
      quality: 100

  Varanda_da_Entrada:
    ffmpeg:    
      inputs:
        - path: rtsp://127.0.0.1:8554/Varanda_da_Entrada
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Varanda_da_Entrada_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: Varanda_da_Entrada_Sub
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25
      stationary:
        interval: 0
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    objects:
      track:
        - person
        - bird
        - cat
        - dog
        - knife
        - cell phone
        - mouse
      mask: 0,0,1000,0,1000,200,0,200
      filters:
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.7
          mask: 0,0,1000,0,1000,200,0,200
    motion:
      threshold: 25
      contour_area: 30
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 50
      mask: 0,900,1080,900,1080,1920,0,1920
      improve_contrast: False
    best_image_timeout: 60
    # Optional: 24/7 recording configuration
    record:
      # Optional: Enable recording (default: global setting)
      enabled: True
      # Optional: Number of days to retain (default: global setting)
      retain:
        days: 3
        mode: motion
      events:
        retain:
          default: 7
          mode: active_objects
          objects:
            person: 50
    # Optional: RTMP re-stream configuration
    rtmp:
      # Required: Enable the live stream (default: True)
      enabled: False
    # Optional: Configuration for the jpg snapshots written to the clips directory for each event
    snapshots:
      # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
      # This value can be set via MQTT and will be updated in startup based on retained value
      enabled: False
      # Optional: print a timestamp on the snapshots (default: shown below)
      timestamp: False
      # Optional: draw bounding box on the snapshots (default: shown below)
      bounding_box: False
      # Optional: crop the snapshot (default: shown below)
      crop: False
      # Optional: height to resize the snapshot to (default: original size)
      height: 175
      # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
      required_zones: []
      # Optional: Camera override for retention settings (default: global values)
      retain:
        # Required: Default retention days (default: shown below)
        default: 10
        # Optional: Per object retention days
        objects:
          person: 15
    mqtt:
      enabled: True
      timestamp: True
      bounding_box: True
      crop: True
      height: 500
      quality: 100

  Lateral_Quartos:
    ffmpeg:    
      inputs:
        - path: rtsp://127.0.0.1:8554/Lateral_Quartos
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Lateral_Quartos_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: Lateral_Quartos_Sub
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25
      stationary:
        interval: 0
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    objects:
      track:
        - person
        - bird
        - cat
        - dog
        - knife
        - cell phone
        - mouse
      mask: 0,0,1000,0,1000,200,0,200
      filters:
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.7
          mask: 0,0,1000,0,1000,200,0,200
    motion:
      threshold: 25
      contour_area: 30
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 50
      mask: 0,900,1080,900,1080,1920,0,1920
      improve_contrast: False
    best_image_timeout: 60
    # Optional: 24/7 recording configuration
    record:
      # Optional: Enable recording (default: global setting)
      enabled: True
      # Optional: Number of days to retain (default: global setting)
      retain:
        days: 3
        mode: motion
      events:
        retain:
          default: 7
          mode: active_objects
          objects:
            person: 50
    # Optional: RTMP re-stream configuration
    rtmp:
      # Required: Enable the live stream (default: True)
      enabled: False
    # Optional: Configuration for the jpg snapshots written to the clips directory for each event
    snapshots:
      # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
      # This value can be set via MQTT and will be updated in startup based on retained value
      enabled: False
      # Optional: print a timestamp on the snapshots (default: shown below)
      timestamp: False
      # Optional: draw bounding box on the snapshots (default: shown below)
      bounding_box: False
      # Optional: crop the snapshot (default: shown below)
      crop: False
      # Optional: height to resize the snapshot to (default: original size)
      height: 175
      # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
      required_zones: []
      # Optional: Camera override for retention settings (default: global values)
      retain:
        # Required: Default retention days (default: shown below)
        default: 10
        # Optional: Per object retention days
        objects:
          person: 15
    mqtt:
      enabled: True
      timestamp: True
      bounding_box: True
      crop: True
      height: 500
      quality: 100

  Traseira:
    ffmpeg:    
      inputs:
        - path: rtsp://127.0.0.1:8554/Traseira
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Traseira_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: Traseira_Sub
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25
      stationary:
        interval: 0
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    objects:
      track:
        - person
        - bird
        - cat
        - dog
        - knife
        - cell phone
        - mouse
      mask: 0,0,1000,0,1000,200,0,200
      filters:
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.7
          mask: 0,0,1000,0,1000,200,0,200
    motion:
      threshold: 25
      contour_area: 30
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 50
      mask: 0,900,1080,900,1080,1920,0,1920
      improve_contrast: False
    best_image_timeout: 60
    # Optional: 24/7 recording configuration
    record:
      # Optional: Enable recording (default: global setting)
      enabled: True
      # Optional: Number of days to retain (default: global setting)
      retain:
        days: 3
        mode: motion
      events:
        retain:
          default: 7
          mode: active_objects
          objects:
            person: 50
    # Optional: RTMP re-stream configuration
    rtmp:
      # Required: Enable the live stream (default: True)
      enabled: False
    # Optional: Configuration for the jpg snapshots written to the clips directory for each event
    snapshots:
      # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
      # This value can be set via MQTT and will be updated in startup based on retained value
      enabled: False
      # Optional: print a timestamp on the snapshots (default: shown below)
      timestamp: False
      # Optional: draw bounding box on the snapshots (default: shown below)
      bounding_box: False
      # Optional: crop the snapshot (default: shown below)
      crop: False
      # Optional: height to resize the snapshot to (default: original size)
      height: 175
      # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
      required_zones: []
      # Optional: Camera override for retention settings (default: global values)
      retain:
        # Required: Default retention days (default: shown below)
        default: 10
        # Optional: Per object retention days
        objects:
          person: 15
    mqtt:
      enabled: True
      timestamp: True
      bounding_box: True
      crop: True
      height: 500
      quality: 100

  Lateral_WC:
    ffmpeg:    
      inputs:
        - path: rtsp://127.0.0.1:8554/Lateral_WC
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Lateral_WC_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: Lateral_WC_Sub
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25
      stationary:
        interval: 0
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    objects:
      track:
        - person
        - car
        - motorcycle
        - bird
        - cat
        - dog
        - knife
        - cell phone
        - mouse
      mask: 0,0,1000,0,1000,200,0,200
      filters:
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.7
          mask: 0,0,1000,0,1000,200,0,200
    motion:
      threshold: 25
      contour_area: 30
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 50
      mask: 0,900,1080,900,1080,1920,0,1920
      improve_contrast: False
    best_image_timeout: 60
    # Optional: 24/7 recording configuration
    record:
      # Optional: Enable recording (default: global setting)
      enabled: True
      # Optional: Number of days to retain (default: global setting)
      retain:
        days: 3
        mode: motion
      events:
        retain:
          default: 7
          mode: active_objects
          objects:
            person: 50
    # Optional: RTMP re-stream configuration
    rtmp:
      # Required: Enable the live stream (default: True)
      enabled: False
    # Optional: Configuration for the jpg snapshots written to the clips directory for each event
    snapshots:
      # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
      # This value can be set via MQTT and will be updated in startup based on retained value
      enabled: False
      # Optional: print a timestamp on the snapshots (default: shown below)
      timestamp: False
      # Optional: draw bounding box on the snapshots (default: shown below)
      bounding_box: False
      # Optional: crop the snapshot (default: shown below)
      crop: False
      # Optional: height to resize the snapshot to (default: original size)
      height: 175
      # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
      required_zones: []
      # Optional: Camera override for retention settings (default: global values)
      retain:
        # Required: Default retention days (default: shown below)
        default: 10
        # Optional: Per object retention days
        objects:
          person: 15
    mqtt:
      enabled: True
      timestamp: True
      bounding_box: True
      crop: True
      height: 500
      quality: 100

  Lateral_Rampa:
    ffmpeg:    
      inputs:
        - path: rtsp://127.0.0.1:8554/Lateral_Rampa
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Lateral_Rampa_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: Lateral_Rampa_Sub
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25
      stationary:
        interval: 0
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    objects:
      track:
        - person
        - car
        - motorcycle
        - bird
        - cat
        - dog
        - knife
        - cell phone
        - mouse
      mask: 0,0,1000,0,1000,200,0,200
      filters:
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.7
          mask: 0,0,1000,0,1000,200,0,200
    motion:
      threshold: 25
      contour_area: 30
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 50
      mask: 0,900,1080,900,1080,1920,0,1920
      improve_contrast: False
    best_image_timeout: 60
    # Optional: 24/7 recording configuration
    record:
      # Optional: Enable recording (default: global setting)
      enabled: True
      # Optional: Number of days to retain (default: global setting)
      retain:
        days: 3
        mode: motion
      events:
        retain:
          default: 7
          mode: active_objects
          objects:
            person: 50
    # Optional: RTMP re-stream configuration
    rtmp:
      # Required: Enable the live stream (default: True)
      enabled: False
    # Optional: Configuration for the jpg snapshots written to the clips directory for each event
    snapshots:
      # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
      # This value can be set via MQTT and will be updated in startup based on retained value
      enabled: False
      # Optional: print a timestamp on the snapshots (default: shown below)
      timestamp: False
      # Optional: draw bounding box on the snapshots (default: shown below)
      bounding_box: False
      # Optional: crop the snapshot (default: shown below)
      crop: False
      # Optional: height to resize the snapshot to (default: original size)
      height: 175
      # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
      required_zones: []
      # Optional: Camera override for retention settings (default: global values)
      retain:
        # Required: Default retention days (default: shown below)
        default: 10
        # Optional: Per object retention days
        objects:
          person: 15
    mqtt:
      enabled: True
      timestamp: True
      bounding_box: True
      crop: True
      height: 500
      quality: 100

  Frente_Portao:
    ffmpeg:    
      inputs:
        - path: rtsp://127.0.0.1:8554/Frente_Portao
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Frente_Portao_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: Frente_Portao_Sub
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25
      stationary:
        interval: 0
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    objects:
      track:
        - person
        - car
        - motorcycle
        - bird
        - cat
        - dog
        - knife
        - cell phone
        - mouse
      mask: 0,0,1000,0,1000,200,0,200
      filters:
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.7
          mask: 0,0,1000,0,1000,200,0,200
    motion:
      threshold: 25
      contour_area: 30
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 50
      mask: 0,900,1080,900,1080,1920,0,1920
      improve_contrast: False
    best_image_timeout: 60
    # Optional: 24/7 recording configuration
    record:
      # Optional: Enable recording (default: global setting)
      enabled: True
      # Optional: Number of days to retain (default: global setting)
      retain:
        days: 3
        mode: motion
      events:
        retain:
          default: 7
          mode: active_objects
          objects:
            person: 50
    # Optional: RTMP re-stream configuration
    rtmp:
      # Required: Enable the live stream (default: True)
      enabled: False
    # Optional: Configuration for the jpg snapshots written to the clips directory for each event
    snapshots:
      # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
      # This value can be set via MQTT and will be updated in startup based on retained value
      enabled: False
      # Optional: print a timestamp on the snapshots (default: shown below)
      timestamp: False
      # Optional: draw bounding box on the snapshots (default: shown below)
      bounding_box: False
      # Optional: crop the snapshot (default: shown below)
      crop: False
      # Optional: height to resize the snapshot to (default: original size)
      height: 175
      # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
      required_zones: []
      # Optional: Camera override for retention settings (default: global values)
      retain:
        # Required: Default retention days (default: shown below)
        default: 10
        # Optional: Per object retention days
        objects:
          person: 15
    mqtt:
      enabled: True
      timestamp: True
      bounding_box: True
      crop: True
      height: 500
      quality: 100

  Campainha:
    ffmpeg:    
      inputs:
        - path: rtsp://127.0.0.1:8554/Campainha
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/Campainha_Sub
          input_args: preset-rtsp-restream
          roles:
            - detect
    live:
      stream_name: Campainha_Sub
    detect:
      width: 1920
      height: 1080
      fps: 5
      enabled: True
      max_disappeared: 25
      stationary:
        interval: 0
        threshold: 50
        max_frames:
          default: 3000
          objects:
            person: 1000
    objects:
      track:
        - person
        - car
        - motorcycle
        - bird
        - cat
        - dog
        - knife
        - cell phone
        - mouse
      mask: 0,0,1000,0,1000,200,0,200
      filters:
        person:
          min_area: 5000
          max_area: 100000
          min_score: 0.5
          threshold: 0.7
          mask: 0,0,1000,0,1000,200,0,200
    motion:
      threshold: 25
      contour_area: 30
      delta_alpha: 0.2
      frame_alpha: 0.2
      frame_height: 50
      mask: 0,900,1080,900,1080,1920,0,1920
      improve_contrast: False
    best_image_timeout: 60
    # Optional: 24/7 recording configuration
    record:
      # Optional: Enable recording (default: global setting)
      enabled: True
      # Optional: Number of days to retain (default: global setting)
      retain:
        days: 3
        mode: motion
      events:
        retain:
          default: 7
          mode: active_objects
          objects:
            person: 50
    # Optional: RTMP re-stream configuration
    rtmp:
      # Required: Enable the live stream (default: True)
      enabled: False
    # Optional: Configuration for the jpg snapshots written to the clips directory for each event
    snapshots:
      # Optional: Enable writing jpg snapshot to /media/frigate/clips (default: shown below)
      # This value can be set via MQTT and will be updated in startup based on retained value
      enabled: False
      # Optional: print a timestamp on the snapshots (default: shown below)
      timestamp: False
      # Optional: draw bounding box on the snapshots (default: shown below)
      bounding_box: False
      # Optional: crop the snapshot (default: shown below)
      crop: False
      # Optional: height to resize the snapshot to (default: original size)
      height: 175
      # Optional: Restrict snapshots to objects that entered any of the listed zones (default: no required zones)
      required_zones: []
      # Optional: Camera override for retention settings (default: global values)
      retain:
        # Required: Default retention days (default: shown below)
        default: 10
        # Optional: Per object retention days
        objects:
          person: 15
    mqtt:
      enabled: True
      timestamp: True
      bounding_box: True
      crop: True
      height: 500
      quality: 100

It’s all working.
My issues are:

  • Low quality for detection, I already changed my streaming detection to primary stream, but frigate crashed with too much cpu usage (95%), despite frigate detects very well if is person, car, dog, cat and whatsoever.
  • Compreface recognizes faces with very low quality, and gives false positives. One example, compreface always gives a result for everyone, even if I only have my family in database. Everyone looks like me or my kids!

Can you help me configuring this better?
Thanks!

You’re not using valid hwaccel args as the docs recommend. The 704x576 should be good for objects that are not far away.

If you want to use a 1080p resolution (which in frigate 0.11 uses the CPU to resize) then I would suggest moving to the frigate 0.12 beta and using ffmpeg presets Hardware Acceleration | Frigate for hardware acceleration which will reduce CPU usage considerably as the resizing of the stream will be done on the GPU instead

You need to configure doubletake to use the correct size images. By default they are downscaled from frigate which leads to poor results. You also need to configure doubletake which % confidences are considered a potential true positive otherwise you will see the behavior where faces are always given a name.

I have doubletake configured with compreface and it is extremely accurate, but of course I only consider detections with 90%+ to be true positive and only consider values of 80%+ to be unknown (everything with lower scores are removed immediately)

If you want to use a 1080p resolution (which in frigate 0.11 uses the CPU to resize) then I would suggest moving to the frigate 0.12 beta

I’m already using it. I’m using 0.12.0-27A31E7 version

You’re not using valid hwaccel args as the docs recommend

Wich args should I use? Intel-based CPUs (<10th Generation) via VAAPI? Intel-based CPUs (>=10th Generation) via Quicksync?

You need to configure doubletake to use the correct size images.

You also need to configure doubletake which % confidences are considered a potential true positive otherwise you will see the behavior where faces are always given a name

How to do that?

Okay, then your hwaccel args should be preset-vaapi

it is all covered in the docs GitHub - jakowenko/double-take: Unified UI and API for processing and training images for facial recognition.

But should I use only the primary stream with frigate, as the secondary one is low quality?

depends on your expectations, using the sub stream will use less resources but maybe won’t perform as well depending on your camera view

I use the sub stream for detection and it is only 704x576. Isn’t that too low for detection?

Not necessarily, all of this is explained in the docs Camera setup | Frigate