[Detector Support]: nvidia GPU vs. Google Coral #15688
-
Describe the problem you are havingHi team, as a big fan of Frigate and Frigate+ subscriber I truely love it :) Frigate runs on my HPE Microserver with Intel Xeon 2,8 Ghz CPU + 32 GB RAM + NVIDIA T 400 4GB Until now, I use Google Coral and just saw, I could use yolonas instead of mobiledet. My question is, based on the above hardware, is it better to switch from Coral to GPU? :) Thank you for the amazing product and your support. Version0.15.0-87e7b62 Frigate config filemodel:
path: plus://0830f79ca1f72a7049d5d84c96b5a919
mqtt:
enabled: true
host: xxxx
port: 1883
user: hass-mqtt
password: mqtt
detectors:
coral:
type: edgetpu
device: usb
ffmpeg:
hwaccel_args: preset-nvidia
motion:
threshold: 15
improve_contrast: true
contour_area: 15
detect:
enabled: true
width: 1280
height: 720
fps: 10
stationary:
interval: 50
threshold: 50
birdseye:
enabled: true
mode: continuous
record:
enabled: true
retain:
mode: motion
alerts:
retain:
days: 30
pre_capture: 30
post_capture: 30
detections:
retain:
days: 30
pre_capture: 30
post_capture: 30
snapshots:
enabled: true
clean_copy: true
bounding_box: true
crop: false
retain:
default: 30
objects:
person: 30
cat: 30
bicycle: 30
motorcycle: 30
car: 30
cameras:
Klingel:
ffmpeg:
output_args:
record: preset-record-generic-audio-copy
inputs:
- path: rtsp://127.0.0.1:8554/Klingel
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://127.0.0.1:8554/Klingel-Detect
input_args: preset-rtsp-restream
roles:
- detect
objects:
track:
- person
- cat
- bicycle
- motorcycle
- car
filters:
person:
min_score: 0.75
threshold: 0.85
min_area: 5000
min_ratio: 0
max_ratio: 0.8
car:
min_score: 0.80
threshold: 0.86
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.8
mask:
- 0.477,0.626,0.784,0.664,0.894,0.624,0.916,0.509,0.477,0.464
- 0.265,0.668,0.28,0.608,0.255,0.529,0,0.471,0,0.603
bicycle:
min_score: 0.75
threshold: 0.85
motorcycle:
min_score: 0.75
threshold: 0.85
cat:
min_score: 0.80
threshold: 0.90
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
# dog:
# min_score: 0.60
# threshold: 0.75
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
review:
alerts:
labels:
- person
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
detections:
labels:
- cat
- car
- bicycle
- motorcycle
- bark
- meow
Einfahrt:
ffmpeg:
output_args:
record: preset-record-generic-audio-copy
inputs:
- path: rtsp://127.0.0.1:8554/Einfahrt
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://127.0.0.1:8554/Einfahrt-Detect
input_args: preset-rtsp-restream
roles:
- detect
- audio
audio:
enabled: true
listen:
- bark
- meow
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
objects:
track:
- person
- cat
- bicycle
- motorcycle
- car
filters:
person:
min_score: 0.75
threshold: 0.75
min_area: 5000
min_ratio: 0
max_ratio: 0.8
car:
min_score: 0.75
threshold: 0.75
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.8
mask:
- 0.076,0.25,0.449,0.166,0.445,0.493,0,0.522,0,0.272
- 0.008,0.612,0.153,0.642,0.231,0.671,0.847,0.617,1,1,0.003,0.997
- 0.587,0,0.604,0.223,0.915,0.196,0.92,0.036
bicycle:
min_score: 0.75
threshold: 0.85
motorcycle:
min_score: 0.75
threshold: 0.85
cat:
min_score: 0.80
threshold: 0.85
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
# dog:
# min_score: 0.60
# threshold: 0.80
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
review:
alerts:
labels:
- person
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
detections:
labels:
- cat
- car
- bicycle
- motorcycle
- bark
- meow
Eingang:
ffmpeg:
output_args:
record: preset-record-generic-audio-copy
inputs:
- path: rtsp://127.0.0.1:8554/Eingang
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://127.0.0.1:8554/Eingang-Detect
input_args: preset-rtsp-restream
roles:
- detect
- audio
audio:
enabled: true
listen:
- bark
- meow
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
objects:
track:
- person
- cat
- bicycle
- motorcycle
- car
filters:
person:
min_score: 0.70
threshold: 0.75
min_area: 5000
min_ratio: 0
max_ratio: 0.8
car:
min_score: 0.70
threshold: 0.75
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.8
mask:
- 0.199,0.059,0.201,0,0.533,0,0.744,0,0.827,0,1,0,1,0.356,1,0.628,0.766,0.494,0.583,0.376,0.431,0.27,0.382,0.165,0.353,0.17
- 0.34,0.519,0.457,0.095,0.719,0.115,1,0.235,0.98,0.956
bicycle:
min_score: 0.70
threshold: 0.75
motorcycle:
min_score: 0.70
threshold: 0.75
cat:
min_score: 0.70
threshold: 0.80
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
# dog:
# min_score: 0.60
# threshold: 0.75
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
review:
alerts:
labels:
- person
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
detections:
labels:
- cat
- car
- bicycle
- motorcycle
- bark
- meow
Garten:
ffmpeg:
output_args:
record: preset-record-generic-audio-copy
inputs:
- path: rtsp://127.0.0.1:8554/Garten
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://127.0.0.1:8554/Garten-Detect
input_args: preset-rtsp-restream
roles:
- detect
- audio
audio:
enabled: true
listen:
- bark
- meow
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
objects:
track:
- person
- cat
filters:
person:
min_score: 0.70
threshold: 0.80
min_area: 5000
min_ratio: 0
max_ratio: 0.8
cat:
min_score: 0.65
threshold: 0.65
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
# dog:
# min_score: 0.60
# threshold: 0.75
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
mask:
0.264,0.691,0.551,0.464,0.674,0.411,0.855,0.478,0.837,0.88,0.797,1,0.17,1,0.175,0.836
review:
alerts:
labels:
- person
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
detections:
labels:
- cat
- car
- bicycle
- motorcycle
- bark
- meow
Haus:
ffmpeg:
output_args:
record: preset-record-generic-audio-copy
inputs:
- path: rtsp://127.0.0.1:8554/Haus
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://127.0.0.1:8554/Haus-Detect
input_args: preset-rtsp-restream
roles:
- detect
- audio
audio:
enabled: true
listen:
- bark
- meow
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
objects:
track:
- person
- cat
filters:
person:
min_score: 0.70
threshold: 0.75
min_area: 2000
min_ratio: 0
max_ratio: 0.8
cat:
min_score: 0.70
threshold: 0.70
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
# dog:
# min_score: 0.60
# threshold: 0.75
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
mask: 0.548,0.407,0.665,0.399,0.675,0.522,0.552,0.54,0.541,0.505,0.544,0.465
review:
alerts:
labels:
- person
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
detections:
labels:
- cat
- car
- bicycle
- motorcycle
- bark
- meow
Garage:
ffmpeg:
output_args:
record: preset-record-generic-audio-copy
inputs:
- path: rtsp://127.0.0.1:8554/Garage
input_args: preset-rtsp-restream
roles:
- record
- path: rtsp://127.0.0.1:8554/Garage-Detect
input_args: preset-rtsp-restream
roles:
- detect
- audio
audio:
enabled: true
listen:
- bark
- meow
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
objects:
track:
- person
- cat
filters:
person:
min_score: 0.80
threshold: 0.90
min_area: 5000
min_ratio: 0
max_ratio: 0.8
cat:
min_score: 0.80
threshold: 0.90
# min_area: 5000
# min_ratio: 1.25
# max_ratio: 1.9
review:
alerts:
labels:
- person
- alarm
- fire_alarm
- car_alarm
- scream
- crying
- speech
- yell
detections:
labels:
- cat
- car
- bicycle
- motorcycle
- bark
- meow
go2rtc:
streams:
Klingel:
- ffmpeg:rtsp://xx@xx/mpeg/media.amp#video=copy#audio=copy#audio=opus
Klingel-Detect:
- ffmpeg:rtsp://xx@xx/mpeg/media.amp#video=copy#audio=copy#audio=opus
Einfahrt:
- ffmpeg:rtsp:/xx@xx/livestream/11#video=copy#audio=copy#audio=opus
Einfahrt-Detect:
- ffmpeg:rtsp://xx@xx/livestream/12#video=copy#audio=copy#audio=opus
Eingang:
- ffmpeg:rtsp://xx@xx/livestream/11#video=copy#audio=copy#audio=opus
Eingang-Detect:
- ffmpeg:rtsp://xx@xx/livestream/12#video=copy#audio=copy#audio=opus
Haus:
- ffmpeg:rtsp://xx@xx/livestream/11#video=copy#audio=copy#audio=opus
Haus-Detect:
- ffmpeg:rtsp://xx@xx/livestream/12#video=copy#audio=copy#audio=opus
Garten:
- ffmpeg:rtsp://xx@xx/livestream/11#video=copy#audio=copy#audio=opus
Garten-Detect:
- ffmpeg:rtsp://xx@xx/livestream/12#video=copy#audio=copy#audio=opus
Garage:
- ffmpeg:rtsp://xx@xx/livestream/11#video=copy#audio=copy#audio=opus
Garage-Detect:
- ffmpeg:rtsp://xx@xx/livestream/12#video=copy#audio=copy#audio=opus
version: 0.15-0
camera_groups:
Vorderseite:
order: 1
icon: LuArrowBigUp
cameras:
- Einfahrt
- Eingang
- Klingel
Rückseite:
order: 2
icon: LuArrowBigDown
cameras:
- Garten
- Garage
- Hausdocker-compose file or Docker CLI command-Relevant Frigate log output-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.
Answered by
NickM-27
Dec 26, 2024
Replies: 1 comment 1 reply
-
|
There's no easy or blanket answer. Blake said in tests the models perform similarly, in my personal experience the yolonas model on nvidia GPU does detect some far away / small objects better at the cost of more resource usage and power usage. Many users have reported similar performance as well. |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
Thanks a lot - I will check it out and will train a model and compare myself, thanks a lot :) |
Beta Was this translation helpful? Give feedback.
Answer selected by
sebastian-tischer
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
There's no easy or blanket answer. Blake said in tests the models perform similarly, in my personal experience the yolonas model on nvidia GPU does detect some far away / small objects better at the cost of more resource usage and power usage. Many users have reported similar performance as well.