[Config Support]: Face recognition is not working at all

[hauserpro]

Describe the problem you are having

Hi, I upgraded to Frigate version: 0.16-0 and enabled the face recognition by using the face_recognition:
enabled: true
model_size: large

However when I browse to see the recognised faces this is what i see:
"There are no recent face recognition attempts"

I have attached the config, can anyone point out what I am missing please? TIA

Version

0.16.0-1fa7ce5

Frigate config file

############################################################
#############      GLOBAL    SETTINGS          #############
############################################################
version: 0.16-0

#############
### MQTT ####
#############
mqtt:
  host: x.x.x.x
  user: user
  password: password

#############
### FFMPEG ##
#############
ffmpeg:
  hwaccel_args: preset-vaapi
  input_args: preset-rtsp-generic
  output_args:
    record: preset-record-generic-audio-aac

#################
### DETECTORS ###
#################
detectors:
  coral:
    type: edgetpu
    device: pci:0
  # coral1:
  #   type: edgetpu
  #   device: usb
  # coral2:
  #   type: edgetpu
  #   device: pci:1


###############
### OBJECTS ###
###############
objects:
  track:
    - person
    # - car
    # - truck
    - bicycle
    - motorcycle
    - dog
    - cat
    - bird
    - umbrella
  filters:
    person:
      min_area: 5000
      max_area: 100000


#################
### RECORDING ###
#################
record:
  enabled: true
  retain:
    days: 7
    mode: all
  alerts:
    retain:
      days: 60
      mode: motion
  detections:
    retain:
      days: 60
      mode: motion


############################################################
#############      CAMERA    SETTINGS          #############
############################################################


###########  CAMERA  ############
cameras:


######################
### Front Doorbell ###
######################
  Frigate-Front-Doorbell:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_09_main
          roles:
            - record
        # Low Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_09_sub
          roles:
            - detect
    detect:
      width: 896
      height: 512
      fps: 10
    zones:
      Zone_Front_Door_Patio:
        coordinates: 0.709,0.717,0.47,0.783,0.493,0.998,0.893,0.999
        loitering_time: 0
    motion:
      mask:
        - 0.659,0,0.664,0.118,0.894,0.125,0.905,0.569,0.779,0.578,0.659,0.677,0.657,0.781,0.766,0.905,0.835,1,0.944,0.995,0.997,0.994,1,0
        - 0.001,0.437,0.363,0.503,0.486,0.634,0.533,0.765,0.528,0.998,0.003,0.999
    review:
      alerts:
        required_zones: Zone_Front_Door_Patio




#########################
### Driveway Doorbell ###
#########################
  Frigate-Driveway-Doorbell:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_10_main
          roles:
            - record
        # Low Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_10_sub
          roles:
            - detect
    detect:
      width: 896
      height: 512
      fps: 10
    zones:
      Zone_Driveway_Doorbell:
        coordinates: 0.004,0.005,0.569,0.222,0.577,0.495,0.992,0.585,0.996,0.997,0.004,0.998
        loitering_time: 0
    motion:
      mask: 
        0.679,0.008,0.995,0.006,0.996,0.067,1,0.591,0.681,0.527,0.572,0.531,0.567,0.427,0.555,0.013
    review:
      alerts:
        required_zones: Zone_Driveway_Doorbell




#########################
### Garden Back       ###
#########################
  Frigate-Garden-Back:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_01_main
          roles:
            - record
        # Low Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_01_sub
          roles:
            - detect
    detect:
# This also effects what you see on the main Frigate screen
      width: 896
      height: 512
      fps: 10
    zones:
      Zone_Garden_Back:
        coordinates: 0,0,0.725,0.006,0.743,0.207,0.754,0.359,1,0.375,0.998,0.995,0.003,0.994
        loitering_time: 0
        inertia: 3
        objects:
          - person
          - umbrella
    motion:
      mask:
        - 0.571,0,0.571,0.095,0.73,0.099,0.74,0.359,1,0.383,1,0
        - 0.513,0.192,0.723,0.209,0.729,0.295,0.516,0.294
        - 0,0.615,0.114,0.622,0.133,0.992,0.006,0.992
        - 0.171,0.532,0.257,0.533,0.257,0.646,0.177,0.663
    review:
      alerts:
        required_zones: Zone_Garden_Back



#########################
### House Front Left  ###
#########################
  Frigate-House-Front-Left:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_05_main
          roles:
            - record
        # Low Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_05_sub
          roles:
            - detect
    detect:
      width: 896
      height: 512
      fps: 10
    zones:
      Zone_House_Front_Left:
        coordinates: 
          0.581,0.992,0.832,0.992,0.978,0.374,0.953,0.298,0.881,0.305,0.837,0.348,0.8,0.378,0.777,0.406,0.752,0.451,0.672,0.632,0.615,0.772,0.577,0.884,0.56,0.961
        loitering_time: 0
    motion:
      mask: 
        0.001,0.002,1,0,1,0.603,0.791,0.483,0.758,0.446,0.656,0.663,0.866,0.854,0.832,1,0,1
    review:
      alerts:
        required_zones: Zone_House_Front_Left




#########################
### Car Right     #######
#########################
  Frigate-Car-Right:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_08_main
          roles:
            - record
        # Low Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_08_sub
          roles:
            - detect
    detect:
      width: 896
      height: 512
      fps: 10
    zones:
      Zone_Car_Right:
        coordinates: 0,0.078,0.957,0.08,0.934,0.406,0.996,0.441,0.995,0.989,0.005,0.992
        loitering_time: 0
    motion:
      mask:
        - 0.571,0.004,0.999,0.005,0.997,0.449,0.924,0.402,0.951,0.088,0.571,0.087
        - 0,0.708,0.227,0.606,0.163,0.193,0,0.287
    review:
      alerts:
        required_zones: Zone_Car_Right


#########################
### Car Left      #######
#########################
    objects:
      mask: 0.626,0.296,0.818,0.597,0.656,0.879,0.433,0.489
  Frigate-Car-Left:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_04_main
          roles:
            - record
            - audio
        # Low Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_04_sub
          roles:
            - detect
    detect:
      width: 896
      height: 512
      fps: 5
    zones:
      Zone_Car_Left:
        coordinates: 0.061,0.442,0.241,0.407,0.492,0.371,0.812,0.384,0.795,1,0,1,0,0.483
        loitering_time: 0
        inertia: 3
    motion:
      mask: 
        0.001,0,0.997,0.008,0.997,0.993,0.798,0.994,0.81,0.39,0.488,0.38,0.066,0.445,0,0.487
    review:
      alerts:
        required_zones: Zone_Car_Left


#########################
### Garden Right  #######
#########################
  Frigate-Garden-Right:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_06_main
          roles:
            - record
            - audio
        # Low Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_06_sub
          roles:
            - detect
    detect:
      width: 896
      height: 512
      fps: 5
    zones:
      Zone_Garden_Right:
        coordinates: 0.003,0.113,0.506,0.016,0.645,0.043,0.693,0.096,0.786,0.102,1,0.252,1,1,0,1
        loitering_time: 0
        inertia: 3
        objects:
          - umbrella
          - person
          - motorcycle
          - dog
          - cat
    motion:
      mask:
        - 0.003,0,0.997,0,0.997,0.25,0.788,0.101,0.695,0.098,0.644,0.04,0.502,0.034,0.114,0.133,0,0.165
        - 0.101,0.559,0.159,0.557,0.187,0.677,0.127,0.709
    review:
      alerts:
        required_zones: Zone_Garden_Right



#########################
### Garage        #######
#########################
  Frigate-Garage:
    ffmpeg:
      inputs:
        # High Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_02_main
          roles:
            - record
        # Low Resolution Stream
        - path: rtsp://admin:xxxxx@10.0.0.10:554/h264Preview_02_sub
          roles:
            - detect
    detect:
      width: 896
      height: 512
      fps: 5
    zones:
      Zone_Garege:
        coordinates: 0.006,0.09,0.997,0.118,0.994,0.994,0.006,0.991
        loitering_time: 0
    review:
      alerts:
        required_zones: Zone_Garege
    motion:
      mask:
        - 0.682,0.027,0.984,0.027,0.985,0.102,0.685,0.105
        - 0.001,0.26,0.19,0.165,0.194,0.197,0.003,0.296
        - 0.246,0.139,0.438,0.086,0.45,0.123,0.25,0.179
        - 0.268,0.521,0.465,0.404,0.474,0.465,0.287,0.592
    objects:
      filters:
        bicycle:
          mask: 0.029,0.479,0.174,0.453,0.244,0.826,0.009,0.912,0,0.497
detect:
  enabled: true
semantic_search:
  enabled: false
  model_size: small
face_recognition:
  enabled: true
  model_size: large
lpr:
  enabled: true
classification:
  bird:
    enabled: false

Relevant Frigate log output

unknown | 2025-06-19 17:00:10 | unknown                               | You have successfully opted out to send the telemetry data.
info    | 2025-06-19 17:00:10 | startup                               | Preparing Frigate...
info    | 2025-06-19 17:00:10 | startup                               | Starting Frigate...
info    | 2025-06-19 17:00:12 | frigate.util.config                   | Checking if frigate config needs migration...
info    | 2025-06-19 17:00:12 | frigate.util.config                   | copying config as backup...
info    | 2025-06-19 17:00:12 | frigate.util.config                   | Finished frigate config migration...
info    | 2025-06-19 17:00:12 | frigate.app                           | Starting Frigate (0.16.0-1fa7ce5)
info    | 2025-06-19 17:00:12 | peewee_migrate.logs                   | Starting migrations
info    | 2025-06-19 17:00:12 | peewee_migrate.logs                   | There is nothing to migrate
info    | 2025-06-19 17:00:12 | frigate.app                           | Recording process started: 341
info    | 2025-06-19 17:00:12 | frigate.app                           | Review process started: 352
info    | 2025-06-19 17:00:12 | frigate.app                           | go2rtc process pid: 115
info    | 2025-06-19 17:00:12 | detector.coral                        | Starting detection process: 363
info    | 2025-06-19 17:00:12 | frigate.app                           | Embedding process started: 365
unknown | 2025-06-19 17:00:12 | unknown                               | INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
info    | 2025-06-19 17:00:12 | frigate.detectors.plugins.edgetpu_tfl | Attempting to load TPU as pci:0
info    | 2025-06-19 17:00:12 | frigate.detectors.plugins.edgetpu_tfl | TPU found
info    | 2025-06-19 17:00:12 | frigate.app                           | Output process started: 414
info    | 2025-06-19 17:00:13 | frigate.app                           | Camera processor started for Frigate-Front-Doorbell: 530
info    | 2025-06-19 17:00:13 | frigate.app                           | Camera processor started for Frigate-Driveway-Doorbell: 537
info    | 2025-06-19 17:00:13 | frigate.app                           | Camera processor started for Frigate-Garden-Back: 545
info    | 2025-06-19 17:00:13 | frigate.app                           | Camera processor started for Frigate-House-Front-Left: 557
info    | 2025-06-19 17:00:13 | frigate.app                           | Camera processor started for Frigate-Car-Right: 566
info    | 2025-06-19 17:00:13 | frigate.app                           | Camera processor started for Frigate-Car-Left: 583
info    | 2025-06-19 17:00:13 | frigate.app                           | Camera processor started for Frigate-Garden-Right: 589
info    | 2025-06-19 17:00:13 | frigate.app                           | Camera processor started for Frigate-Garage: 600
info    | 2025-06-19 17:00:13 | frigate.app                           | Capture process started for Frigate-Front-Doorbell: 604
info    | 2025-06-19 17:00:13 | frigate.app                           | Capture process started for Frigate-Driveway-Doorbell: 625
info    | 2025-06-19 17:00:13 | frigate.app                           | Capture process started for Frigate-Garden-Back: 642
info    | 2025-06-19 17:00:13 | frigate.app                           | Capture process started for Frigate-House-Front-Left: 652
unknown | 2025-06-19 17:00:13 | unknown                               | loading data from : /config/model_cache/facedet/landmarkdet.yaml
info    | 2025-06-19 17:00:13 | frigate.app                           | Capture process started for Frigate-Car-Right: 665
info    | 2025-06-19 17:00:13 | frigate.app                           | Capture process started for Frigate-Car-Left: 675
info    | 2025-06-19 17:00:13 | frigate.app                           | Capture process started for Frigate-Garden-Right: 682
info    | 2025-06-19 17:00:13 | frigate.app                           | Capture process started for Frigate-Garage: 698
info    | 2025-06-19 17:00:23 | frigate.api.fastapi_app               | Starting FastAPI app
info    | 2025-06-19 17:00:23 | frigate.api.fastapi_app               | FastAPI started

Relevant go2rtc log output

please see the full config above.

Frigate stats

No response

Operating system

Proxmox

Install method

Docker Compose

docker-compose file or Docker CLI command

services:
  frigate:
    container_name: frigate
    privileged: true
    restart: unless-stopped
    #image: ghcr.io/blakeblackshear/frigate:stable
    image: ghcr.io/blakeblackshear/frigate:0.16.0-beta2
    shm_size: "800mb" # update for your cameras based on calculation above
    devices:
      - /dev/bus/usb:/dev/bus/usb
      - /dev/dri/renderD128 # for intel hwaccel, needs to be updated for your hardware
    volumes:
      - /etc/localtime:/etc/localtime:ro
      - /root/docker:/config
      - /cctv_clips:/media/frigate
      - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
        target: /tmp/cache
        tmpfs:
          size: 1000000000
    ports:
      - "8971:8971"
      - "5000:5000" # Internal unauthenticated access. Expose carefully.
      - "8554:8554" # RTSP feeds
      - "8555:8555/tcp" # WebRTC over tcp
      - "8555:8555/udp" # WebRTC over udp
      - "1935:1935" # RTMP feeds
    environment:
      FRIGATE_RTSP_PASSWORD: "password"

Object Detector

Coral

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

No response

[NickM-27]

Have you created any face collections yet?

[hauserpro]

sorry, no. I thought all we had to do was enable the face recognition using the command?
face_recognition:
enabled: true

Can you please point me in direction to the manual which shows how to do that?

[hawkeye217]

https://deploy-preview-16390--frigate-docs.netlify.app/configuration/face_recognition#usage

[hauserpro]

Oh boy, i feel so silly now. So all I had to do was add some images lol.

Thank you for pointing this out. Once I added some images it kicked into life and started seeing some faces on face recognition page.

[hauserpro]

Before we close this I noticed only one camera is picking up faces for recognition. Do I need to enable it per camera?

[hawkeye217]

If you enabled it globally or with the switch in the UI, then no, you don't need to enable it per camera.

You'll want to review the debugging steps in the docs: https://deploy-preview-16390--frigate-docs.netlify.app/configuration/face_recognition/#how-do-i-debug-face-recognition-issues

[molnarti]

i am also struggling to figure out how this is supposed to work. i have enabled face recognition in the settings, set up some faces with sample images for training. but:

• detection is running on the sub stream which is 896 x 672 big.
• on the substream the size of the WHOLE person (not just his face) is approx 100 * 120 pixels
• the default minimal threshold for face detection is 500 pixels

So is this aimed for cameras such as doorbells where you expect someone is directly looking into the camera? Because i see no way how to detect faces reliably from the substream on a camera mounted 3 meters high

[hawkeye217]

No, you should choose a detect resolution that fits your specific scene and requirements.

See the official documentation: https://docs.frigate.video/configuration/face_recognition

Also, see the 0.16 FAQ, Usage question 7.

If you still have issues, you should open your own support discussion.

[NickM-27]

To be clear, the entire person is 12,000 pixels in area in your example. The face min area is 500, this seems fine, I'm not sure why this is being pointed out as an issue

[kanocz]

you can use my fork... PR is not accepted, but I'll try to keep up to date #22529

[alexsmartens]

I hit the same issue and spent hours debugging "Detected no faces for person object" with various stream resolutions, model sizes, and thresholds before finding this thread.

The root cause was simply that the Face Library was empty. Once I uploaded a single face, detection immediately started working.

The current UX is misleading because:

1. The Face Library page says "There are no recent face recognition attempts" which implies the system is waiting/working but nothing has happened yet, not that it's inactive due to missing training data

2. The debug log message "Detected no faces for person object" suggests a detection quality issue (resolution, angle, etc.) rather than a prerequisite not being met

3. The Settings page lets you enable face recognition without any indication that uploading a face is required before the pipeline activates

A simple improvement could be a notice on the Face Library page when it's empty, something like: Upload at least one face to activate face detection. Face recognition will not scan for faces until the library contains at least one entry.

[NickM-27]

yes, an empty card has been implemented for the next version

[hawkeye217]

Worth mentioning that the official docs also indicate the need to upload at least one face: https://docs.frigate.video/configuration/face_recognition/#usage

[alexsmartens]

Fair point, it is listed in the Usage steps. But I think the issue is that the docs describe it as a sequential workflow ("follow these steps to begin"), which is easy to skim past when you're troubleshooting.

The real UX gap is what happens between step 1 and step 2: you enable face recognition, restart, see person detections working, but the Face Library shows "no recent face recognition attempts" with zero indication of why. The natural assumption is that something is broken and not that the system is waiting for you to complete step 2 before it starts.

A prominent warning note (red background, like the existing NOTE and TIP callout boxes but more urgent) right after the config section would help:

IMPORTANT: Face detection will not run until at least one face has been uploaded to the Face Library. If your Face Library shows "no recent face recognition attempts," ensure you have completed step 2 below.

The upcoming empty state card in the UI will also help a lot!

[hawkeye217]

The upcoming empty state card in the UI will also help a lot!

Definitely!

[celesteking]

It's still unclear why it isn't labeling faces in "person" snapshots under /explore tab. I see it labels them as "person", but where's the face name I've already trained in "face library"?

All these objects persons labels tags collections recognizions snapshots clips alerts detections motions, all this is extremely and utterly confusing. Outrageously complicated for something that was supposed to be so simple.

Can't you make a 3 min video on how it is supposed to look like or write a short article? Docs are terrible - no screenshots, no explanations.

[NickM-27]

When a person is correctly recognized, the face name will show in explore (along with a button to edit it if you need to)

[celesteking]

It took that from /classification of a person object. And that classification was created initially when the library was (almost?) empty -- I was so perplexed on what to do next after the install that I was just poking the UI.
It's very confusing and no walk-through exists on what to do next after the cams have been added.

Why isn't it adding a separate "Faces" category under /explore? All Labels filter at the top-right doesn't list faces I've configured.

[NickM-27]

It's very confusing and no walk-through exists on what to do next after the cams have been added.

Frigate can be used for a myriad of different use case, for any given use case some features will be useful and others will have no reason to exist. We can't add a walkthrough for every type of use case from residential to businesses. Each feature has a dedicated page in the docs which explain how it works and how to use it.

Frigate 0.18 added full UI configuration, and with that the docs have been updated to also include some explanations on each primary screen and how to use it.

It's very confusing and no walk-through exists on what to do next after the cams have been added.

How would we know what you want to do next? Some users don't care about face recognition at all, for example. This is something that users decide for themselves, it will naturally be easier when the full config is in the UI, but "what to do next" is a question based on the goals and use case of the individual.

Why isn't it adding a separate "Faces" category under /explore? All Labels filter at the top-right doesn't list faces I've configured.

Explore is for searching for objects, face is not an object, it is an attribute of a person. If you want to filter on recognized people then you use the More Filters dropdown and select Resident for example.

[celesteking]

Frigate 0.18 added full UI configuration, and with that the docs have been updated to also include some explanations on each primary screen and how to use it.

I should've waited for the release.

It's very confusing and no walk-through exists on what to do next after the cams have been added.
How would we know what you want to do next?

It's called a DEMO. Where you demonstrate all available functions of your product for the user to decide whether he wants to buy it or not. Despite this being open-source, you just show them what's available. And, as this is an UI driven software (or is trying to become one), you put up a video or, well, a working demo. The one at https://demo.frigate.video/ doesn't include /classification page. Also, what's up with the left bar that can't expand to show the text labels? I'm an old man, I can't understand pictograms. See https://m3.material.io/components/navigation-rail/overview

Why isn't it adding a separate "Faces" category under /explore? All Labels filter at the top-right doesn't list faces I've configured.

Explore is for searching for objects, face is not an object, it is an attribute of a person. If you want to filter on recognized people then you use the More Filters dropdown and select Resident for example.

I believe you misunderstood my complaint. "Resident" isn't a face I've configured in /faces. It's a classifier I had configured in /classification. I had (have) no idea what I've been doing. The idea ("what to do next") was to assign a "resident" label to block tenants (aka the "person" objects, per frigate terminology). In addition to that I now want to recognize their faces so that it labels them both as "resident" and as their face name. Currently, it doesn't do that.

As re /explore functionality, there's a magnifying glass icon and it's called "Explore". I want to do exactly that. I want to look up (filter) a person by their name. I sincerely don't care whether it's not an object or whatever, it's none of my business. I also don't understand why this limitation exists.

I don't understand the line of thinking behind coming up with such complicated terminology. Why can't everything just be an object with infinite number of tags? In docs and UI, I came across classes, objects, attributes, labels, sub-labels, models. It's too convoluted. The deeper I dive in, the more frustrated I get.

Thanks for replying.

[NickM-27]

There's no conceivable way that we could have a demo for every feature and every way it can be used. Just like when you download windows or Linux, you have to figure some things out for yourself based on how you want to use it.

I want to look up (filter) a person by their name

Like it says in the docs you can't configure classification and face recognition for a person at the same time, they will conflict. To do this you should setup face recognition.

I don't understand the line of thinking behind coming up with such complicated terminology. Why can't everything just be an object with infinite number of tags?

It doesn't make sense for a face or license plate (or UPS logo, FedEx logo, etc) to be a normal object, it's a sub type on top of person, car, etc. Frigate tracks primary objects with metadata stored about that object, like the face for a given person as you'd want these to be associated for filtering reasons.

Frigate is only just starting to become easier to use for non-technical users, there is a long way to go and it won't make sense for our guides to be built out until we land somewhere that won't change every version.

It sounds like there's probably software that will better suite what you're looking for.