[Detector Support]: Frigate+ 2024.3 detection oddity #15520
-
Describe the problem you are havingConsistantly since trying 2024.3, several detections seem to have been mapped incorrectly. Most notable, 'licence_plate' gets detected as 'cat' and 'ups' gets detected as 'bear'. Another oddity is 'squirrel' gets detected as 'bird' pretty regularly but also sometimes 'umbrella', so not sure about this one. I thought this was due to low training #s but wasn't sure until suggestions in frigate+ were available. The suggestions are spot on every time. I've been using the tensorrt docker image with a P4000, a P2000 and 2x minipcii coral bridge things from here. Not sure multiple corals is related to the problem. Happy to put things in debug for logs if needed, just need to know how. Version0.15.0-869fa26 Frigate config filemqtt:
host: mqtt
database:
path: /db/frigate.db
detectors:
coral_pci0:
type: edgetpu
device: pci:0
coral_pci1:
type: edgetpu
device: pci:1
coral_pci2:
type: edgetpu
device: pci:2
coral_pci3:
type: edgetpu
device: pci:3
model:
path: plus://[PLUSCODE]
semantic_search:
enabled: true
model_size: large
reindex: false
birdseye:
enabled: true
restream: false
mode: continuous
quality: 5
detect:
stationary:
interval: 5000
threshold: 45
objects:
track:
- bbq_grill
- bear
- bicycle
- bird
- boat
- car
- cat
- cow
- deer
- dhl
- dog
- face
- fedex
- fox
- goat
- horse
- license_plate
- motorcycle
- package
- person
- rabbit
- raccoon
- robot_lawnmower
- squirrel
- umbrella
- ups
- usps
- waste_bin
audio:
enabled: true
listen:
- bark
- fire_alarm
- scream
- speech
- yell
- bang
- glass
- gunshot
- explosion
- boom
- siren
- door
- doorbell
- lawn_mower
- police_car
- ambulance
- fire_engine
- emergency_vehicle
- honk
record:
enabled: true
retain:
days: 14
alerts:
retain:
days: 14
pre_capture: 10
post_capture: 10
detections:
retain:
days: 14
pre_capture: 10
post_capture: 10
snapshots:
enabled: true
retain:
default: 14
objects:
person: 30
car: 30
go2rtc:
streams:
Doorbell:
- rtspx://192.168.7.1:7441/yGZZwZ0jUMzda7h1
Doorbell_package:
- rtspx://192.168.7.1:7441/XyS63nB06PWEtgJd
webrtc:
candidates:
- 192.168.7.201:8555
- stun:8555
cameras:
Garage:
enabled: true
birdseye:
mode: objects
objects:
track:
- person
- cat
- dog
- bird
- face
ffmpeg:
input_args: preset-rtsp-restream
output_args:
record: preset-record-generic-audio-aac
inputs:
- path: rtsp://viewonly:viewonly12@192.168.7.181:554/cam/realmonitor?channel=1&subtype=0
roles:
- record
- detect
- audio
motion:
mask: 0.816,0.043,0.816,0.073,0.969,0.071,0.969,0.043
FrontDoor:
enabled: true
motion:
threshold: 40
contour_area: 30
improve_contrast: true
ffmpeg:
inputs:
- path: rtsp://127.0.0.1:8554/Doorbell
roles:
- record
- detect
- audio
output_args:
record: preset-record-ubiquiti
objects: {}
FrontDoor_PackageCam:
enabled: true
ffmpeg:
inputs:
- path: rtsp://127.0.0.1:8554/Doorbell_package
roles:
- record
- detect
- audio
FrontDeck:
enabled: true
objects:
track:
- car
- person
- face
- cat
- dog
- bird
ffmpeg:
input_args: preset-rtsp-restream
output_args:
record: preset-record-generic-audio-aac
inputs:
- path: rtsp://viewonly:viewonly12@192.168.7.182:554/cam/realmonitor?channel=1&subtype=0
roles:
- record
- detect
- audio
motion:
mask: 0.813,0.072,0.972,0.072,0.971,0.041,0.813,0.04
BackYard:
enabled: true
ffmpeg:
input_args: preset-rtsp-restream
output_args:
record: preset-record-generic-audio-aac
inputs:
- path: rtsp://viewonly:viewonly12@192.168.7.183:554/cam/realmonitor?channel=1&subtype=0
roles:
- record
- detect
- audio
motion:
mask: 0.816,0.041,0.817,0.071,0.97,0.074,0.969,0.042
Driveway:
enabled: true
ffmpeg:
input_args: preset-rtsp-restream
output_args:
record: preset-record-generic-audio-aac
inputs:
- path: rtsp://viewonly:viewonly12@192.168.7.184:554/cam/realmonitor?channel=1&subtype=0
roles:
- record
- detect
- audio
motion:
mask: 0.014,0.943,0.015,0.974,0.165,0.973,0.165,0.945
threshold: 40
contour_area: 30
improve_contrast: true
version: 0.15-0 docker-compose file or Docker CLI commandservices:
frigate:
container_name: frigate
# add groups from host for render, plugdev, video
group_add:
- "998" # render
- "39" # video
shm_size: "16384mb"
image: ghcr.io/blakeblackshear/frigate:869fa26-tensorrt
deploy:
resources:
reservations:
devices:
- driver: nvidia
# device_ids: ['0']
capabilities: [gpu]
count: all
devices:
- "/dev/apex_0:/dev/apex_0"
- "/dev/apex_1:/dev/apex_1"
- "/dev/apex_2:/dev/apex_2"
- "/dev/apex_3:/dev/apex_3"
- "/dev/dri/:/dev/dri/"
environment:
- PGID=${PGID}
- PUID=${PUID}
- TZ=${TZ}
- NVIDIA_DRIVER_CAPABILITIES=all
- NVIDIA_VISIBLE_DEVICES=all
- PLUS_API_KEY=${FRIGATE_PLUS_API_KEY}
- YOLO_MODELS=${FRIGATE_YOLO_MODELS}
- TRT_MODEL_PREP_DEVICE=0
- change=true
logging:
driver: json-file
options:
max-file: ${DOCKERLOGGING_MAXFILE}
max-size: ${DOCKERLOGGING_MAXSIZE}
privileged: true
ports:
- '5000:5000/tcp'
- '5001:5001/tcp'
- '8554:8554'
- '8971:8971'
volumes:
- /etc/localtime:/etc/localtime:ro
- /dev/bus/usb:/dev/bus/usb
- ${DOCKERCONFDIR}/frigate/config:/config
- ${DOCKERCONFDIR}/frigate/db/:/db/
- /bigstorage/frigatemedia:/media/frigate
mqtt:
container_name: mqtt
image: eclipse-mosquitto:1.6
ports:
- "1883:1883" Relevant Frigate log outputNo errors relevant in log, Can provide detector debug if I can figure out how to show that. Install methodDocker Compose Object DetectorCoral Screenshots of the Frigate UI's System metrics pagesAny other information that may be helpfulNo response |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments
-
In my searches initially, I thought this might be related to #15421 but I am not using the yolo model. Maybe has something to do with that though. Not sure of the inner workings here. |
Beta Was this translation helpful? Give feedback.
-
You should be applying the model globally not only on the last detector. This is a config issue. |
Beta Was this translation helpful? Give feedback.
-
Ah this is a case of carrying my config from 0.12 through everything until now. I had simply pasted the coral detectors in place where the old tensorrt section was when I got them and the indent lived on. I moved the 'model:' section above the detectors and there have been no licence_plate cats in the last 10min. Thanks for the extra set of eyes Nick! Sorry about the bother. Working config snippet:
|
Beta Was this translation helpful? Give feedback.
Ah this is a case of carrying my config from 0.12 through everything until now.
I had simply pasted the coral detectors in place where the old tensorrt section was when I got them and the indent lived on.
I moved the 'model:' section above the detectors and there have been no licence_plate cats in the last 10min.
Thanks for the extra set of eyes Nick! Sorry about the bother.
Working config snippet: