are/why are "badgers" being ignored?

[ElectronicBattle]

Hello

short version:
Is the latest Frigate+ (F+) model now so accurate, when used in conjunction with 0.17, such that when it "sees" something which clearly isn't a fox or a dog, it will ignore the animal even though previously a badger would have been (mis) labelled?

Detail:
On 0.16.4 using the previous and current (16/1/26) Frigate+ edgetpu model I found that visiting badgers were being (mis)detected as "dog" or "fox" and so I was simply correcting them when submitting to F+ with a "preemptive" candidate label, ready for if/when badgers get classified by F+:

---

---------------------------------------------------------------------------------------------------------------------

Having u/g to version: 0.17-0 I think I am finding that badgers are no longer being (wrongly) classified and in fact are not being detected at all. Other animals e.g. fox and dog and cat are still working, perhaps as reliably as before but I also consider that I may have noticed a decrease in the frequency of classifications.

I have at the same time used AI to adjust my config significantly for readability, a section of which I show here for brief perusal and comment if anyone can see what I have done poorly.

"badger" isn't yet an actual label, so I deleted the instances of badger from objects: track and from filters and from review:alerts:labels` just to clear out the warning logs. This is when I think I am seeing fewer detections; I am going to add "badger" back in a little while.

I have got myself quite confused with the dozens of configs I have worked through, building from my functional baseline to the somewhat more streamlined version below.

I shall restore badger once again and try it but I wonder if 0.17 and the F+ model is so accurate that it is now ignoring animals which it can see are not dogs and are not foxes and hence actual badgers are being ignored.

# --- OBJECT DETECTION & SEMANTIC SEARCH ---
detect:
  enabled: true
  width: 640
  height: 360

objects:
  track:
    - person
    - face
    - car
    - license_plate
    - cat
    - dog
    - fox
    - bird
    - deer
    - squirrel
    - hedgehog
    - badger # I am about to add this back
    - package
    - bicycle
  filters:
    dog: {min_score: 0.8, threshold: 0.9}
    cat: {min_score: 0.9, threshold: 0.85}
    fox: {min_score: 0.75, threshold: 0.8}
    squirrel: {min_score: 0.65, threshold: 0.8}
    bird: {min_score: 0.65, threshold: 0.8}
    hedgehog: {min_score: 0.65, threshold: 0.8}
    badger: {min_score: 0.65, threshold: 0.8} # # I am about to add this back
    face: {min_score: 0.7}
    package: {min_score: 0.65, threshold: 0.9}
    amazon: {min_score: 0.75}
    ups: {min_score: 0.75}
    fedex: {min_score: 0.75}
    person: {min_score: 0.65, threshold: 0.85}
    car: {min_score: 0.65, threshold: 0.85}
    license_plate: {min_score: 0.6}
    bicycle: {min_score: 0.65, threshold: 0.85}

semantic_search:
  enabled: true
  reindex: false
  model_size: small # CPU processing for embedding

# --- RECORDING & SNAPSHOTS ---
record:
  enabled: true
  continuous:
    days: 0  # SSD optimization: No continuous recording of "nothing"
  motion:
    days: 2  # 48hr safety net for non-object motion
  alerts:
    retain:
      days: 7  # High-priority (People/Cars) for 1 week
      mode: motion
  detections:
    retain:
      days: 3  # Low-priority (Wildlife) for 3 days
      mode: motion

snapshots:
  enabled: true
  clean_copy: true
  timestamp: false
  bounding_box: true
  crop: false
  height: 768
  quality: 85
  retain:
    default: 5
    objects:
      person: 7
      face: 7
      cat: 7
      dog: 7
      fox: 7
      badger: 7 # I am about to add this back
      car: 7
      bird: 3
      squirrel: 3

# --- CAMERAS ---
cameras:
  upper_patio: 
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://192.168.1.123:554/media=0/channel=1&level=1/track0
          roles: [detect]
        - path: rtsp://192.168.1.123:554/media=0/channel=1&level=0/track0
          roles: [record]
    motion:
      mask: 
        0.003,0.374,0.207,0.313,0.239,0.272,0.292,0.258,0.32,0.245,0.36,0.222,0.682,0.219,0.716,0.204,0.726,0.001,0.002,0.001
    zones:
      upperpatio:
        coordinates: 
          0.003,0.997,0.005,0.795,0.042,0.789,0.045,0.732,0.049,0.619,0.077,0.546,0.103,0.479,0.145,0.425,0.168,0.401,0.204,0.371,0.216,0.369,0.268,0.378,0.279,0.339,0.32,0.302,0.347,0.293,0.361,0.298,0.382,0.345,0.395,0.375,0.563,0.418,0.582,0.474,0.626,0.476,0.674,0.51,0.712,0.515,0.795,0.664,0.862,0.704,0.933,0.822,0.999,0.896,0.995,0.996
    review:
      alerts:
        required_zones: upperpatio
      detections:
        required_zones: upperpatio
    birdseye:
      order: 2

# --- UI & POST-PROCESSING ---
ui:
  time_format: 24hour

birdseye:
  enabled: true
  mode: continuous
  inactivity_threshold: 5

camera_groups:
  Woodlands2:
    order: 1
    icon: LuActivitySquare
    cameras: [drive, upper_patio, doorbell]
  birdseye:
    order: 2
    icon: LuAmpersands
    cameras: birdseye
  Sitting_room:
    order: 3
    icon: LuAlertCircle
    cameras: doorbell

motion:
  improve_contrast: true

face_recognition:
  enabled: false
  model_size: small

lpr:
  enabled: true

classification:
  bird:
    enabled: true

review:
  alerts:
    labels: [person, face, car, license_plate, cat, dog, fox, deer, hedgehog, badger # I am about to add this back
          bicycle]
  detections:
    labels: [bird, squirrel]

[inkeep-ai-assistant[bot]]

badger is a candidate label in Frigate+, which means it doesn't have enough training data to be included in the model yet. (1)

Candidate labels are available for annotation to help add support sooner, but they are not available for automatic suggestions. (1)

Where possible, candidate labels are mapped to existing labels during training—for example, baby labels are mapped to person. (1) However, there's no indication that badger is currently mapped to dog or fox during inference.

Why badgers may no longer be detected:

The behavior you're observing—where badgers were previously misclassified as dog or fox but are now ignored entirely—likely reflects improved model accuracy. A more accurate model will have higher confidence distinguishing between known classes and will score unknown animals (like badgers) lower, causing them to fall below your min_score and threshold filters. (1)

Your configuration considerations:

• Adding badger to objects: track won't cause Frigate to detect badgers, since the model doesn't support that label yet. (1)
• Your dog filter has a high min_score: 0.8 and threshold: 0.9, which may filter out lower-confidence detections that previously caught badgers as misclassified dogs. (2)

What you can do:

1. Continue submitting badger images to Frigate+ with the badger candidate label to help accelerate support for that label. (1)
2. If you want to catch badgers as misclassified animals in the meantime, consider lowering your dog and fox thresholds slightly to capture more borderline detections. (1)

📚 Sources:

1. Models
2. Full Reference Config

---

Have another question? Just tag @inkeep.