[Bug]: Face detection with negative score : Unable to handle embeddings request too many values to unpack

[gmarti]

Checklist

• I have updated to the latest available Frigate version.
• I have cleared the cache of my browser.
• I have tried a different browser to see if it is related to my browser.
• I have tried reproducing the issue in incognito mode to rule out problems with any third party extensions or plugins I have installed.
• I have asked the AI at https://docs.frigate.video about my issue.

Describe the problem you are having

Reprocess a face fails with "Unable to handle embeddings request too many values to unpack"

Happens mostly with new faces or with a new installation.

The names in clips/faces/train are wrongly named with double dash:

The split fails because there is 6 instead of 5 values https://github.com/blakeblackshear/frigate/blob/v0.16.1/frigate/data_processing/real_time/face.py#L402

The problem is most likely because write_face_attempt get a negative score

Steps to reproduce

1. New installation with only one face trained
2. Record multiple person with new faces
3. Somehow the faces files get wrongly named
4. Try to reprocess face => it fails

Version

0.16.1

In which browser(s) are you experiencing the issue with?

Chromium 140.0.7339.127

Frigate config file

auth:
  enabled: true
cameras:
  doorbell:
    detect:
      enabled: true
      height: 1920
      width: 2560
      fps: 5
    ffmpeg:
      inputs:
        - input_args: preset-rtsp-restream-low-latency
          path: rtsp://127.0.0.1:8554/doorbell
          roles:
            - record
            - detect
    live:
      streams:
        doorbell: doorbell
    motion:
      contour_area: 10
      improve_contrast: true
      mask: 0,0,0,0.25,0.455,0.474,0.611,0.515,0.653,0.512,1,0.497,1,0
      threshold: 25
    snapshots:
      bounding_box: true
      enabled: true
      retain:
        default: 3
        objects:
          person: 7
      timestamp: true
detectors:
  coral0:
    device: pci:0
    type: edgetpu
  coral1:
    device: pci:1
    type: edgetpu
face_recognition:
  enabled: true
  model_size: large
ffmpeg:
  hwaccel_args: preset-intel-qsv-h264
model:
  path: plus://
mqtt:
  enabled: false
objects:
  filters:
    car:
      min_score: 0.65
      threshold: 0.85
    cat:
      min_score: 0.65
      threshold: 0.8
    dog:
      min_score: 0.7
      threshold: 0.9
    face:
      min_score: 0.5
      threshold: 0.7
    package:
      min_score: 0.65
      threshold: 0.9
    person:
      min_score: 0.65
      threshold: 0.85
  track:
    - person
    - face
    - dog
    - cat
    - fox
    - badger
    - car
    - deer
    - bird
    - package
    #- waste_bin
    - hedgehog
    - bus
    - tractor
record:
  alerts:
    post_capture: 10
    pre_capture: 5
    retain:
      days: 30
      mode: active_objects
  detections:
    post_capture: 10
    pre_capture: 5
    retain:
      days: 30
      mode: active_objects
  enabled: true
  retain:
    days: 5
    mode: motion
detect:
  enabled: true
go2rtc:
  streams:
    doorbell:
      - rtsp://user:passowrd@ip/cam/realmonitor?channel=1&subtype=0&unicast=true&proto=Onvif#backchannel=0
version: 0.16-0

docker-compose file or Docker CLI command

podman systemd

Relevant Frigate log output

Oct 13 09:45:39 SERVER-01 frigate[578642]: 2025-10-13 09:45:39.071764288  10.125.0.134 - - [13/Oct/2025:09:45:39 +0200] "POST /api/faces/reprocess HTTP/1.1" 200 2 "https://10.125.0.12:8971/faces" "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/140.0.0.0 Safari/537.36" "-"
Oct 13 09:45:39 SERVER-01 frigate[578642]: 2025-10-13 09:45:39.092857443  10.125.0.134 - - [13/Oct/2025:09:45:39 +0200] "GET /api/faces HTTP/1.1" 200 5897 "https://10.125.0.12:8971/faces" "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/140.0.0.0 Safari/537.36" "-"
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056737112  [2025-10-13 09:45:41] frigate.embeddings.maintainer  ERROR   : Unable to handle embeddings request too many values to unpack (expected 5)
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056742944  Traceback (most recent call last):
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056745949    File "/opt/frigate/frigate/embeddings/maintainer.py", line 222, in _handle_request
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056748197      resp = processor.handle_request(topic, data)
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056750148             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056752501    File "/opt/frigate/frigate/data_processing/real_time/face.py", line 402, in handle_request
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056754563      (id_time, id_rand, timestamp, _, _) = current_file.split("-")
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056756265      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Oct 13 09:45:41 SERVER-01 frigate[578642]: 2025-10-13 09:45:41.056758202  ValueError: too many values to unpack (expected 5)

Relevant go2rtc log output

-

Operating system

Other Linux

Install method

Docker CLI

Network connection

Wired

Camera make and model

Amcrest AD410

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[inkeep-ai-assistant[bot]]

The issue you're experiencing appears to be related to how Frigate handles face recognition file naming when faces receive negative scores. Based on the error message and the file naming pattern you've shown, this is a bug where the face recognition system is creating filenames with additional dashes that break the expected parsing format.

The face recognition system expects face files to follow a specific naming convention for processing (1). When the system encounters faces with negative scores, it appears to be generating filenames that don't conform to the expected 5-part format (separated by dashes), causing the "too many values to unpack" error during reprocessing.

For face recognition to work properly, you need to ensure that person is being detected first, as Frigate needs to detect a person before it can detect and recognize a face (1). The face recognition configuration in your setup looks correct with enabled: true and model_size: large (2).

To troubleshoot this issue:

1. Check that persons are being detected consistently before face recognition attempts
2. Verify your face recognition configuration matches the documented format (2):

face_recognition:
  enabled: true
  model_size: large

(2)

3. Consider adjusting the detection_threshold if you're experiencing issues with face detection quality (1)
4. You may need to clear the corrupted face files from /media/frigate/clips/faces/train and retrain your faces

The face recognition system is designed to save face recognition attempts for training purposes, with a default of 100 saved attempts (1). If the file naming is corrupted due to negative scores, this can prevent proper reprocessing of faces.

This appears to be a software bug that should be reported to the Frigate development team, as the face recognition system should handle negative scores gracefully without breaking the file naming convention.

📚 Sources:

1. Face Recognition
2. Full Reference Config

---

Have another question? Just tag @inkeep.

[NickM-27]

Thanks, I believe this has already been fixed for 0.17

[gmarti]

Are you sure @NickM-27 ?
I compared the code at 0.16.1 and dev and didn't find any changes that would fix this