[Detector Support]: Coral TPU Recognized but Frigate fails to see - Debian 12 Bare Metal #15648
Replies: 2 comments 15 replies
-
|
You're trying to use the openvino model with the coral which is incorrect |
Beta Was this translation helpful? Give feedback.
-
|
Sorry! I had the Coral detector commented out when I copy/pasted the config file as the container will get into a restart loop while the Coral detector is in play. The log output and the compose file are correct from when I had the Coral detector lines in place and not commented out. |
Beta Was this translation helpful? Give feedback.
-
|
Sure, but did you comment out the |
Beta Was this translation helpful? Give feedback.
-
|
It's not clear what detector / model config you actually tried to use |
Beta Was this translation helpful? Give feedback.
-
|
I did not comment out the model when pasting this exact log. I have been leaving the OpenVINO detector in play, as well as the model. I also have tried changing the model to a Plus model i have built that supports Coral but received the same error. I do know this Plus model is not supported on Coral, but expected it to work with OpenVINO, and just leave the Coral idle. I do have a Plus model built for Coral as well that I am happy to try (with no OpenVINO detector in play), or any other steps that might be useful to narrow it down. |
Beta Was this translation helpful? Give feedback.
-
|
I'd suggest commenting the model out entirely and only keeping the coral part, then paste the updated config and logs here |
Beta Was this translation helpful? Give feedback.
-
|
No problem! Below is the updated config: mqtt:
enabled: true
host: 192.168.2.50
port: 1883
user: sebastian
password: Kojack10
tls:
enabled: false
detectors:
# ov:
# type: openvino
# device: GPU
coral:
type: edgetpu
device: pci
# deepstack:
# api_url: http://192.168.2.75:32168/v1/vision/detection
# type: deepstack
# api_timeout: 0.1 # seconds
#model:
#width: 300
#height: 300
#input_tensor: nhwc
#input_pixel_format: bgr
# path: plus://36d1c0614ecdaaad533afced32252ed6
go2rtc:
streams:
doorbell_main:
- rtsp://admin:Kojack10!@192.168.2.234:554/h264Preview_01_main
- ffmpeg:doorbell_main#audio=opus # transcodes audio to opus for webrtc compatibility
doorbell_sub:
- rtsp://admin:Kojack10!@192.168.2.234:554/h264Preview_01_sub
front_main:
- rtsp://admin:Kojack10@192.168.2.134:554/h264Preview_01_main
- ffmpeg:front_main#audio=aac
front_sub: rtsp://admin:Kojack10@192.168.2.134:554/h264Preview_01_sub
front_door_main:
- rtsp://admin:Kojack10@192.168.2.222:554/h264Preview_01_main
- ffmpeg:front_door_main#audio=aac
front_door_sub: rtsp://admin:Kojack10@192.168.2.222:554/h264Preview_01_sub
back_main:
- rtsp://admin:Kojack10@192.168.2.133:554/h264Preview_01_main
- ffmpeg:back_main#audio=aac
back_sub: rtsp://admin:Kojack10@192.168.2.133:554/h264Preview_01_sub
homekit:
doorbell_main: # same stream ID from streams list, default PIN - 19550224
front_main:
front_door_main:
back_main:
ffmpeg:
# hwaccel_args: -c:v h264_qsv
hwaccel_args: preset-vaapi
# hwaccel_args: -hwaccel vaapi -hwaccel_device /dev/dri/renderD129≈≈
audio: # <- enable audio events for all camera
enabled: true
listen:
- bark
- fire_alarm
- scream
- speech
- yell
cameras:
doorbell: # <------ Name the camera
enabled: true
ffmpeg:
input_args: -rtsp_transport tcp
output_args:
record: preset-record-generic-audio-aac
inputs:
- path: rtsp://127.0.0.1:8554/doorbell_sub # <----- The stream you want to use for detection
input_args: preset-rtsp-restream
roles:
- detect
- path: rtsp://127.0.0.1:8554/doorbell_main
input_args: preset-rtsp-restream
roles:
- record
- audio
detect:
width: 640
height: 480
fps: 7
live:
stream_name: doorbell_main
objects:
track:
- person
- face
- dog
- bicycle
- car
- amazon
- fedex
- ups
- package
- waste_bin
- deer
- rabbit
filters:
person:
threshold: 0.5
car: {}
bicycle: {}
mask: 0.684,0.021,0.689,0.048,0.983,0.049,0.982,0.021
snapshots:
enabled: true
timestamp: false
bounding_box: true
retain:
default: 30
record:
enabled: true
retain:
days: 3
mode: all
sync_recordings: true
alerts:
retain:
days: 30
pre_capture: 3
detections:
retain:
days: 30
pre_capture: 3
zones:
front_porch:
coordinates: 0.283,0.625,0.574,0.603,0.607,0.624,0.999,0.755,0.955,1,0.224,1,0.288,0.689
inertia: 3
loitering_time: 60
neighbor_driveway:
coordinates: 0.614,0.624,0.751,0.59,0.822,0.626,0.822,0.698
inertia: 3
loitering_time: 0
Street:
coordinates: 0.284,0.624,0.79,0.579,0.789,0.471,0.284,0.485
loitering_time: 0
motion:
# Optional: The percentage of the image used to detect lightning or other substantial changes where motion detection
# needs to recalibrate. (default: shown below)
# Increasing this value will make motion detection more likely to consider lightning or ir mode changes as valid motion.
# Decreasing this value will make motion detection more likely to ignore large amounts of motion such as a person approaching
# a doorbell camera.
lightning_threshold: 0.6
mask:
- 0.684,0.022,0.688,0.048,0.983,0.048,0.983,0.017
- 0,0,0.347,0,1,0,1,0.349,0.282,0.352,0.283,0.356,0,0.342
- 0,0.341,0.283,0.355,0.278,0.522,0.286,0.697,0,1
threshold: 30
contour_area: 10
improve_contrast: 'false'
review:
alerts:
required_zones: front_porch
detections:
required_zones: front_porch
front_yard: # <------ Name the camera
enabled: true
ffmpeg:
inputs:
- path: rtsp://127.0.0.1:8554/front_sub # <----- The stream you want to use for detection
input_args: preset-rtsp-restream
roles:
- detect
- path: rtsp://127.0.0.1:8554/front_main
input_args: preset-rtsp-restream
roles:
- record
- audio
detect:
width: 1536
height: 576
fps: 7
live:
stream_name: front_main
objects:
track:
- person
- face
- dog
- bicycle
- car
- amazon
- fedex
- ups
- package
- waste_bin
- deer
- rabbit
filters:
person:
mask: 0.185,0.751,0.244,0.746,0.246,0.823,0.184,0.837
car:
mask: 0.091,0.583,0.234,0.389,0.338,0.491,0.15,0.846
snapshots:
enabled: true
timestamp: false
bounding_box: true
retain:
default: 30
record:
enabled: true
retain:
days: 3
mode: all
sync_recordings: true
alerts:
retain:
days: 30
pre_capture: 3
detections:
retain:
days: 30
pre_capture: 3
motion:
# Optional: The percentage of the image used to detect lightning or other substantial changes where motion detection
# needs to recalibrate. (default: shown below)
# Increasing this value will make motion detection more likely to consider lightning or ir mode changes as valid motion.
# Decreasing this value will make motion detection more likely to ignore large amounts of motion such as a person approaching
# a doorbell camera.
lightning_threshold: 0.6
mask:
- 0.774,0.016,0.775,0.074,0.991,0.072,0.991,0.016
- 0.024,0.065,0.118,0.065,0.117,0.147,0.024,0.149
- 0.276,-0.002,0.203,0.145,0.139,0.298,0.075,0.457,0.048,1,0,1,0,0
- 0.74,0,0.825,0.153,0.879,0.242,0.928,0.392,0.968,0.475,1,0.543,0.999,0.009
threshold: 39
contour_area: 10
improve_contrast: 'false'
zones:
driveway:
coordinates: 0.424,0.247,0.579,0.254,0.71,1,0.275,1
inertia: 3
loitering_time: 0
street:
coordinates:
0.725,0.309,0.564,0.252,0.425,0.242,0.303,0.283,0.307,0.195,0.406,0.138,0.504,0.125,0.584,0.152,0.662,0.188,0.729,0.233
inertia: 3
loitering_time: 0
review:
alerts:
required_zones:
- driveway
- street
detections:
required_zones:
- street
- driveway
front_door: # <------ Name the camera
enabled: true
ffmpeg:
inputs:
- path: rtsp://127.0.0.1:8554/front_door_main # <----- The stream you want to use for detection
input_args: preset-rtsp-restream
roles:
- record
- audio
- path: rtsp://127.0.0.1:8554/front_door_sub
input_args: preset-rtsp-restream
roles:
- detect
detect:
width: 896
height: 512
fps: 8
live:
stream_name: front_door_main
objects:
track:
- person
- face
- dog
- bicycle
- car
- amazon
- fedex
- ups
- package
- waste_bin
- deer
- rabbit
snapshots:
enabled: true
timestamp: false
bounding_box: true
retain:
default: 30
record:
enabled: true
retain:
days: 3
mode: all
sync_recordings: true
alerts:
retain:
days: 30
pre_capture: 3
detections:
retain:
days: 30
pre_capture: 3
motion:
# Optional: The percentage of the image used to detect lightning or other substantial changes where motion detection
# needs to recalibrate. (default: shown below)
# Increasing this value will make motion detection more likely to consider lightning or ir mode changes as valid motion.
# Decreasing this value will make motion detection more likely to ignore large amounts of motion such as a person approaching
# a doorbell camera.
lightning_threshold: 0.6
improve_contrast: 'false'
mask: 0.577,0.023,0.58,0.072,1,0.069,0.989,0.016
zones:
front_porch:
coordinates: 0.311,0.518,0.535,0.305,0.776,0.781,0.631,0.998,0.504,0.997
inertia: 3
loitering_time: 0
review:
alerts:
required_zones: front_porch
detections:
required_zones: front_porch
back_yard: # <------ Name the camera
enabled: true
ffmpeg:
input_args: -rtsp_transport tcp
inputs:
- path: rtsp://127.0.0.1:8554/back_sub # <----- The stream you want to use for detection
input_args: preset-rtsp-restream
roles:
- detect
- path: rtsp://127.0.0.1:8554/back_main
input_args: preset-rtsp-restream
roles:
- record
- audio
detect:
width: 1536
height: 576
fps: 7
live:
stream_name: back_main
objects:
track:
- person
- dog
filters:
person:
mask: 0.185,0.751,0.244,0.746,0.246,0.823,0.184,0.837
snapshots:
enabled: true
timestamp: false
bounding_box: true
retain:
default: 30
record:
enabled: true
retain:
days: 3
mode: motion
sync_recordings: true
alerts:
retain:
days: 30
pre_capture: 3
detections:
retain:
days: 3
motion:
# Optional: The percentage of the image used to detect lightning or other substantial changes where motion detection
# needs to recalibrate. (default: shown below)
# Increasing this value will make motion detection more likely to consider lightning or ir mode changes as valid motion.
# Decreasing this value will make motion detection more likely to ignore large amounts of motion such as a person approaching
# a doorbell camera.
lightning_threshold: 0.6
mask:
- 0.005,0.063,0.215,0.065,0.212,0.012,0.004,0.014
- 0.329,0.559,0.297,0.688,0.341,0.727,0.37,0.585
threshold: 33
contour_area: 16
improve_contrast: 'false'
zones:
yard:
coordinates:
0.206,0.746,0.406,0.138,0.507,0.105,0.663,0.135,0.801,0.993,0.466,0.998,0.291,0.998
inertia: 3
loitering_time: 0
review:
alerts:
required_zones: yard
detections: {}
version: 0.15-0
camera_groups:
Birdseye:
order: 1
icon: LuHome
cameras: birdseyeAnd below is the updated log: |
Beta Was this translation helpful? Give feedback.
-
|
All of these are run on the host. |
Beta Was this translation helpful? Give feedback.
-
|
need to see the first two inside the container |
Beta Was this translation helpful? Give feedback.
-
|
No problem! ls /dev/ output: lspci output:
|
Beta Was this translation helpful? Give feedback.
-
|
@NickM-27 I just wanted to check back and see if there were any thoughts here? If not, I can continue using the OpenVino model on the CPU as the interference speed is fairly good. It is just interesting that CodeProject will see the TPU with no issue, so I didn't know if there was something I was missing. |
Beta Was this translation helpful? Give feedback.
-
|
only thing I can think is some power issue or something, not clear why it is not working correctly |
Beta Was this translation helpful? Give feedback.
-
Describe the problem you are having
I am on a fresh install of Debian 12 on bare metal. I have installed all the drivers according to the Coral docs for the PCIe card, and have verified that everything shows up correctly in Debian.
I have installed docker and got Frigate all up and running, and everything works according to plan except for the Coral detector. It does not see the device and fails to start.
I am currently trying the v0.15 beta, but this error also occurs on v0.14.
I experimented with installing the CodeProject.ai docker container on the same machine to see if it will detect and process images through the TPU and it does work correctly.
Version
0.15.0-87e7b62
Frigate config file
docker-compose file or Docker CLI command
Relevant Frigate log output
Install method
Docker Compose
Object Detector
Coral
Screenshots of the Frigate UI's System metrics pages
N/A
Any other information that may be helpful
No response
Beta Was this translation helpful? Give feedback.
All reactions