[Support]: Face detection misidentifies person objects as "non person", causing silent failure (0.16.2) #20873
-
Checklist
Describe the problem you are havingFace detection is completely non-functional on Frigate 0.16.2 running on Raspberry Pi 5 with Hailo-8L, despite proper configuration and working license plate recognition. The embeddings manager successfully processes car/motorcycle events for LPR, but silently skips face detection on person events without any errors or warnings. Critical Discovery: Debug Logs Reveal the BugWith debug logging enabled ( The face detection code is receiving person objects but incorrectly identifying them as "non person" objects, causing it to skip face processing entirely. Symptoms
ReproductionReproduced on TWO separate instances:
Issue is NOT configuration-related. Face detection code misidentifies ALL person objects as "non person". How to Reproduce the Debug LogsEnable debug logging: logger:
default: info
logs:
frigate.data_processing.real_time.face: debugThen check logs: docker logs frigate 2>&1 | grep -i "face"You'll see the repeated "Not a processing face for non person object" message. License Plate Recognition Works (Proves Embeddings Manager Functional)"recognized_license_plate": "06/1",
"recognized_license_plate_score": 0.9909341782331467This proves the embeddings manager IS working correctly for car/motorcycle objects, but face detection specifically fails on person objects. Related: This appears to be the same issue as Discussion #19860 Steps to reproduce... Version0.16.2-4d58206 In which browser(s) are you experiencing the issue with?all Frigate config filemqtt:
enabled: false
logger:
default: info
logs:
frigate.data_processing.real_time.face: debug
frigate.embeddings: debug
database:
path: /media/frigate/frigate.db
detectors:
cpu:
type: cpu
face_recognition:
enabled: true
model_size: large
min_area: 200
cameras:
parking:
enabled: true
ffmpeg:
inputs:
- path: rtsp://admin:[email protected]:554/Preview_01_sub
roles:
- detect
- path: rtsp://admin:[email protected]:554/Preview_01_main
roles:
- record
detect:
enabled: true
width: 640
height: 360
fps: 5
objects:
track:
- person
- car
record:
enabled: true
retain:
days: 1
mode: motion
snapshots:
enabled: true
retain:
default: 1
detect:
enabled: true
version: 0.16-0docker-compose file or Docker CLI commandservices:
frigate:
container_name: frigate-test
privileged: true
restart: unless-stopped
image: ghcr.io/blakeblackshear/frigate:stable
shm_size: "256mb"
devices:
- /dev/bus/usb:/dev/bus/usb
- /dev/dri/renderD128:/dev/dri/renderD128
volumes:
- /etc/localtime:/etc/localtime:ro
- ./config:/config
- ./storage:/media/frigate
- type: tmpfs
target: /tmp/cache
tmpfs:
size: 1000000000
ports:
- "5001:5000" # Web UI on different port
- "8556:8554" # RTSP
- "8557:8555/tcp" # WebRTC
- "8557:8555/udp" # WebRTC
environment:
FRIGATE_RTSP_PASSWORD: "password"Relevant Frigate log output# Embeddings manager starts successfully
[2025-11-09 18:45:53] frigate.app INFO : Embedding process started: 446
loading data from : /config/model_cache/facedet/landmarkdet.yaml
# Face models downloaded successfully
[2025-11-09 18:45:54] frigate.util.downloader INFO : Downloading complete:
https://github.com/NickM-27/facenet-onnx/releases/download/v1.0/facedet.onnx
[2025-11-09 18:46:18] frigate.util.downloader INFO : Downloading complete:
https://github.com/NickM-27/facenet-onnx/releases/download/v1.0/landmarkdet.yaml
[2025-11-09 18:47:08] frigate.util.downloader INFO : Downloading complete:
https://github.com/NickM-27/facenet-onnx/releases/download/v1.0/arcface.onnx
# Debug logs showing the bug (with frigate.data_processing.real_time.face: debug)
[2025-11-09 19:27:11] frigate.data_processing.real_time.face DEBUG : Not a processing face for non person object.
[2025-11-09 19:27:17] frigate.data_processing.real_time.face DEBUG : Not a processing face for non person object.
[2025-11-09 19:27:18] frigate.data_processing.real_time.face DEBUG : Not a processing face for non person object.
[2025-11-09 19:27:20] frigate.data_processing.real_time.face DEBUG : Not a processing face for non person object.
[2025-11-09 19:27:20] frigate.data_processing.real_time.face DEBUG : Not a processing face for non person object.
[2025-11-09 19:27:21] frigate.data_processing.real_time.face DEBUG : Not a processing face for non person object.
[2025-11-09 19:27:25] frigate.data_processing.real_time.face DEBUG : Not a processing face for non person object.
[2025-11-09 19:27:27] frigate.data_processing.real_time.face DEBUG : Not a processing face for non person object.
... (repeated hundreds of times for every person detection)
# Note: Person detection is working normally - these messages appear when person objects are detected
# Note: LPR is working successfully on car/motorcycle objects - embeddings manager is functionalRelevant go2rtc log outputN/A - Issue reproduced on test instance with direct RTSP (no go2rtc).
Streaming works fine. Issue is specific to face detection processing.Operating systemDebian Install methodDocker Compose Network connectionWired Camera make and modelReolink Screenshots of the Frigate UI's System metrics pagesAvailable upon request Any other information that may be helpfulKey findings:
Debug logging config used: |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 6 replies
-
|
Based on the provided details and your configuration, Frigate's face recognition is designed to only process faces on objects detected as type Key points for troubleshooting:
Additional checks:
If all configuration is correct and the issue persists across clean installs and multiple detectors (CPU, Hailo-8L), this behavior matches an internal bug, not a configuration or environmental issue. No workaround is documented for this scenario. For further assistance and bug tracking, please visit: Face Recognition Documentation and GitHub. 📚 Sources: Have another question? Just tag @inkeep. |
Beta Was this translation helpful? Give feedback.
-
|
This is not a bug, your assessment of the logs is incorrect. The AI's response is correct - your stream resolution is likely the biggest contributor to the reason why faces are not being detected. You should also try reducing your |
Beta Was this translation helpful? Give feedback.
-
|
● Thank you for the feedback. You were correct - I misinterpreted the debug logs. Changes Made Per Your Suggestions
ResultsDebug logs: Person detection:
Result: Face detection runs on person objects but returns "Detected no faces for person object." I welcome further thoughts. System: Raspberry Pi 5, Frigate 0.16.2-4d58206, Hailo-8L detector |
Beta Was this translation helpful? Give feedback.
-
|
Have you uploaded at least one face? See the documentation: https://docs.frigate.video/configuration/face_recognition/#usage |
Beta Was this translation helpful? Give feedback.
-
|
No, I haven't uploaded any reference faces to the Face Library yet. My understanding from the documentation is that face detection should work independently - unmatched faces should appear in the "Train" tab even without reference faces uploaded. But I'm not seeing any faces detected at all. The logs show: Running manual face detection. Detected no faces for person object. Should I upload a reference face first? Does that change how face detection operates? |
Beta Was this translation helpful? Give feedback.
-
|
That is incorrect. Frigate 0.16 implements face recognition. Without a native Since you're not running a model that detects As the docs indicate, you should upload at least one face to start to see detected faces in the Train tab. The official documentation covers all of this in detail. https://docs.frigate.video/configuration/face_recognition/ |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for your patience, hawkeye217. I apologize for the confusion - I did not understand that uploading at least one reference face was required to jumpstart the face detection/recognition system. After uploading a reference face, recognition is working very well now, even with lower resolution and blurry captures. Issue resolved. Thank you for the help. |
Beta Was this translation helpful? Give feedback.
-
|
Great, glad to hear it's working well for you now. Happy to help! |
Beta Was this translation helpful? Give feedback.
That is incorrect. Frigate 0.16 implements face recognition. Without a native
face-detecting model (either something custom or a model like Frigate+), faces are not tracked and saved as a standard object within Frigate's object tracking system.Since you're not running a model that detects
facenatively, face detection will run via CV2 using a lightweight DNN model that runs on the CPU. Any faces detected through this secondary pipeline are only sent off to the recognition system.As the docs indicate, you should upload at least one face to start to see detected faces in the Train tab.
The official documentation covers all of this in detail. https://docs.frigate.video/configuration/face_r…