[Config Support]: MQTT snapshots do not obey the configured threshold limit

[nacree]

Describe the problem you are having

I subscribed for Frigate+ and meanwhile updated min_score and threshold values in my config to higher numbers according to the docs.

objects:
  filters:
    person:
      min_score: .75
      threshold: .85

I'm following the last detected objects through MQTT in HA dashboard, and now I have noticed that it shows me the objects with score from 75%, even the threshold value is 85%.

I somehow understand the difference between the min_score and threshold values, but does it make sense to increase to min_score value to same value as threshold to drop off the improbable detections from MQTT. Or is there other way to filter out those detections from the MQTT, as they are not then shown on Frigate side at all.

Thanks for the help.

Version

0.16-beta3

Frigate config file

mqtt:
  enabled: true
  host: 192.168.72.8
  client_id: frigate
  port: 1883
  user: .......
  password: ........

tls:
  # Disable TLS for port 8971
  enabled: false

ffmpeg:
  hwaccel_args: preset-intel-qsv-h264
  output_args:
    record: preset-record-generic-audio-aac

detectors:
  ov:
    type: openvino
    device: GPU

model:
  path: plus://f69400e8341d2fe8010271f87d59466e

go2rtc:
  streams:
     ......
     
     
record:
  enabled: true
  retain:
    days: 5
    mode: all
  alerts:
    retain:
      days: 30
  detections:
    retain:
      days: 30

snapshots:
  enabled: true
  bounding_box: true

detect:
  enabled: true
  height: 720
  width: 1280
  fps: 1

live:
  height: 480

objects:
  filters:
    deer:
      min_score: .6
      threshold: .8
    dog:
      min_score: .7
      threshold: .9
    cat:
      min_score: .65
      threshold: .8
    face:
      min_score: .7
    license_plate:
      min_score: .6
    person:
      min_score: .75
      threshold: .85
    car:
      min_score: .75
      threshold: .85

motion:
  enabled: true
  threshold: 30
  mqtt_off_delay: 30
  lightning_threshold: 0.8
  improve_contrast: true

cameras:
  etupiha:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://localhost:8554/etupiha
          roles:
            - record
            - detect

    detect:
      enabled: true
      height: 720
      width: 1280

    record:
      enabled: true
    objects:
      track:
        - person
        - cat
        - dog
        - deer

      filters:
        person:
          mask: 0.887,0.297,0.941,0.289,0.998,0.545,0.912,0.596
    motion:
      enabled: true
      mask: 0.889,0.173,0.902,0,0.001,0.001,0,0.423
      threshold: 45
      contour_area: 30
      improve_contrast: true

  reolink:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://...
          input_args: preset-rtsp-generic
          roles:
            - detect
        - path: rtsp://localhost:8554/reolink
          roles:
            - record
    detect:
      enabled: true
      width: 640
      height: 480
      fps: 5
    record:
      enabled: true
    objects:
      track:
        - person
        - bicycle
        - car
        - cat
        - dog
        - deer
    motion:
      enabled: true
      mask:
        - 0.01,0.016,0.011,0.048,0.305,0.048,0.304,0.016
        - 0.002,0.409,0.355,0.3,0.372,0.176,0.634,0.197,0.745,0.214,0.907,0.222,0.878,0.292,0.908,0.414,0.997,0.48,0.998,0.003,0.001,0.002
      threshold: 40
      contour_area: 20
      improve_contrast: true

  etuovi:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://localhost:8554/etuovi
          roles:
            - record
            - detect
    detect:
      enabled: true
      width: 1152
      height: 648
      fps: 5
    record:
      enabled: true
    objects:
      track:
        - person
        - cat
        - dog
        - deer

      filters:
        person:
          mask: 0.028,0.574,0.123,0.985,0.215,0.933,0.127,0.505
    motion:
      enabled: true
      mask:
        - 0,0,0,1,0.212,1,0.15,0
        - 0.719,0,0.71,0.099,1,0.356,1,0
      threshold: 40
      contour_area: 15
      improve_contrast: true

  keittio:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://localhost:8554/keittioSub
          roles:
            - detect
        - path: rtsp://localhost:8554/keittio
          roles:
            - record
    detect:
      enabled: false
      width: 640
      height: 360
    record:
      enabled: true
    motion:
      enabled: true
      threshold: 40
      mqtt_off_delay: 30
      lightning_threshold: 0.8
      improve_contrast: true
      contour_area: 20
      mask:
        - 0,0.29,0.136,0.766,0.226,0.647,0.108,0.186
        - 0,0.052,0.357,0.053,0.357,0,0,0.001

  piha:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://localhost:8554/piha
          roles:
            - detect
            - record
    detect:
      enabled: true
      width: 1152
      height: 648
      fps: 2
    record:
      enabled: true
    motion:
      enabled: true
      threshold: 35
      contour_area: 15
      improve_contrast: true
    objects:
      track:
        - person
        - cat
        - dog
        - deer

  olkkari:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://localhost:8554/olkkari
          roles:
            - record
    detect:
      enabled: false
      width: 640
      height: 360
    record:
      enabled: true
    motion:
      mask:
        - 0.192,0.272,0.263,0.26,0.358,0.05,0.362,0.002,0,0.002,0.002,0.049
        - 0.833,0.075,0.81,0.294,0.999,0.467,1,0.176
      threshold: 40
      contour_area: 20
      improve_contrast: true

  lastenhuone:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://localhost:8554/lastenhuone
          roles:
            - record
    record:
      enabled: true
    detect:
      enabled: false
      width: 640
      height: 360
    motion:
      enabled: false
      mask: 0,0.046,0.356,0.053,0.358,0.001,0,0.001

  pinnasanky:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://localhost:8554/pinnasanky
          roles:
            - record
    record:
      enabled: true
    detect:
      enabled: false
      width: 640
      height: 360
    motion:
      enabled: false
      mask: 0.001,0.049,0.357,0.053,0.356,0,0.001,0.001

version: 0.16-0
camera_groups:
  Piha:
    order: 1
    icon: LuBird
    cameras:
      - etuovi
      - etupiha
      - piha
      - reolink
  Sisä:
    order: 2
    icon: LuBanana
    cameras:
      - keittio
      - lastenhuone
      - pinnasanky
      - olkkari

Relevant Frigate log output

No relevant log entries

Relevant go2rtc log output

No relevant go2rtc log entries

Frigate stats

No response

Operating system

Debian

Install method

Docker Compose

docker-compose file or Docker CLI command

Not relevant

Object Detector

OpenVino

Screenshots of the Frigate UI's System metrics pages

No response

Any other information that may be helpful

No response

[hawkeye217]

There are a few things to consider here:

• MQTT topics like /events are raw topics without any filters, so they may sometimes have messages for objects that don't appear in the Frigate UI (false positives are one significant example).
• An object's threshold and min_score interact in various ways. See the official documentation.
• Frigate 0.16 beta 4 will include some changes for MQTT snapshots as there were some bugs found in prior betas - specifically relating to zones. Since you didn't post your config, I can't tell you exactly if you're encountering that or not.

[nacree]

@hawkeye217 thanks for the quick response. I have updated my question with the configuration. However I don't have any zones in use, so I doubt that it will bring changes.

[NickM-27]

This is working as expected, the MQTT snapshots don't follow the full rules of the events / review items as users generally want faster response times.

[NickM-27]

Yes, to have low latency image updating. It's consistent with how all the sensors work for the integration

[Mincka]

Would you have any suggestion to have an image entity "filtered with a threshold"?

My goal is to show a widget on my phone with the latest true positive of an object for a camera (or even better for a zone).

Currently, I have image.frigate_front_garden_person, but with the min_score of 0.5 – that I don't want to change – I'm often left with false positive images on my widget.

[NickM-27]

Use /events topic, write the image to a file, use the file integration to have that image shown as an image entity

[Mincka]

Thank you for your guidance @NickM-27, really appreciated.

[Mincka]

For anyone interested, I decided to publish the image directly to MQTT, like this.

Automation to publish the images based on specific events criteria:

alias: Frigate MQTT Update Maison Avant Jardin Avant Person
description: ""
triggers:
  - topic: frigate/events
    id: frigate_maison_avant_jardin_avant_person
    payload: 1/1/1/1/1
    value_template: >-
      {{ iif(value_json["after"]["has_snapshot"] == true, 1) }}/{{
      iif(value_json["after"]["camera"] == "Frigate_Maison_Avant", 1) }}/{{
      iif(value_json["after"]["label"] == "person", 1) }}/{{
      iif(["jardin_avant"]|select("in",
      value_json["after"]["current_zones"])|list|length > 0, 1) }}/{{
      iif(value_json["after"]["top_score"]|float(0) >= 0.86, 1) }}
    variables:
      id: "{{ trigger.payload_json[\"after\"][\"id\"] }}"
    trigger: mqtt
  - topic: frigate/events
    id: frigate_jardin_avant_jardin_avant_person
    payload: 1/1/1/1/1
    value_template: >-
      {{ iif(value_json["after"]["has_snapshot"] == true, 1) }}/{{
      iif(value_json["after"]["camera"] == "Frigate_Jardin_Avant", 1) }}/{{
      iif(value_json["after"]["label"] == "person", 1) }}/{{
      iif(["jardin_avant"]|select("in",
      value_json["after"]["current_zones"])|list|length > 0, 1) }}/{{
      iif(value_json["after"]["top_score"]|float(0) >= 0.86, 1) }}
    variables:
      id: "{{ trigger.payload_json[\"after\"][\"id\"] }}"
    trigger: mqtt
conditions: []
actions:
  - choose:
      - conditions:
          - condition: trigger
            id:
              - frigate_maison_avant_jardin_avant_person
              - frigate_jardin_arriere_jardin_arriere_person
        sequence:
          - action: mqtt.publish
            data:
              qos: "1"
              retain: true
              topic: homeassistant/image/{{ trigger.id }}
              payload: >-
                https://xxx/api/frigate/notifications/{{id}}/thumbnail.jpg
mode: queued
trace:
  stored_traces: 5
max: 10

And in configuration.yml

mqtt:
  - image:
      unique_id: frigate_maison_avant_jardin_avant_person
      name: Frigate Maison Avant - Jardin Avant - Person
      url_topic: homeassistant/image/frigate_maison_avant_jardin_avant_person
  - image:
      unique_id: frigate_jardin_avant_jardin_avant_person
      name: Frigate Jardin Avant - Jardin Avant - Person
      url_topic: homeassistant/image/frigate_jardin_avant_jardin_avant_person