[Detector Support]: TensorRT Crashing [Truenas SCALE] Nvidia GTX 1060 #16674
-
Describe the problem you are havingHi folks. I've been manged getting my GPU drivers working nicely and my GTX 1060 is now pushing hardware accel for decoding perfectly. I really want the GPU in to do detection though as I don't want to bake the CPU and I have no TPU. this is my beautiful nvidia-smi log: what I cant manage to do is to set the GPU as detector, as soon as I define tensorrt object detector Frigate stops working with blank screen, my yaml config is attached, and works fine untill I add: `detectors: model: after addind these last lines it becomes a nightmare, stops working and get blank screen and have to revert changes to the config file removing the detectors stuff.. how can I solve this? regards and tanks Version0.15.0-cea210d Frigate config filemqtt:
enabled: false
detectors:
tensorrt:
type: tensorrt
device: 0 #This is the default, select the first GPU
model:
path: /config/model_cache/tensorrt/yolov7-tiny-320.trt
input_tensor: nchw
input_pixel_format: rgb
width: 320
height: 320
objects:
track:
- person
- bicycle
- motorcycle
- bus
- boat
- bird
- cat
- dog
- frisbee
- sports ball
- kite
- skateboard
- bottle
- cell phone
cameras:
CAM1:
enabled: true
ffmpeg:
inputs:
- path: rtsp://admin:1234567890abcd@192.168.1.198:554/Streaming/Channels/102/
roles:
- detect
- path: rtsp://admin:1234567890abcd@192.168.1.198:554/Streaming/Channels/101/
roles:
- record
- audio
zones:
Fronte_Cancello:
coordinates:
0.158,0.73,0.231,0.775,0.429,0.853,0.668,0.762,0.861,0.665,1,0.599,0.991,0.671,0.953,0.896,0.925,0.992,0.514,0.997,0.119,1,0.059,0.727,0.092,0.705,0.13,0.704
inertia: 3
loitering_time: 3
review:
alerts:
required_zones: Fronte_Cancello
motion:
mask: 0.109,0.944,0.109,0.972,0.386,0.968,0.391,0.944
CAM2:
enabled: true
ffmpeg:
inputs:
- path: rtsp://admin:1234567890abcd@192.168.1.199:554/Streaming/Channels/102/
roles:
- detect
- path: rtsp://admin:1234567890abcd@192.168.1.199:554/Streaming/Channels/101/
roles:
- record
- audio
detect:
enabled: true
zones:
Piazzale:
coordinates:
0.472,0.306,0.473,0.374,0.5,0.372,0.499,0.429,0.77,0.449,0.779,0.536,0.83,0.535,0.827,0.559,0.845,0.559,0.845,0.611,0.892,0.62,0.943,0.629,1,0.63,0.995,1,0.117,0.997,0.044,0.616,0.052,0.285,0.122,0.102,0.283,0.218,0.307,0.263,0.357,0.317
inertia: 3
loitering_time: 0
motion:
mask: 0.108,0.943,0.38,0.942,0.38,0.972,0.127,0.97
review:
alerts:
required_zones: Piazzale
camera_groups:
Piazzale:
order: 2
icon: LuCast
cameras:
- CAM1
- CAM2
record:
enabled: true
retain:
days: 3
mode: motion
semantic_search:
enabled: true
reindex: false
model_size: large
version: 0.15-1docker-compose file or Docker CLI commandghcr.io/blakeblackshear/frigate:0.15.0-tensorrtRelevant Frigate log outputCPU detectors are not recommended and should only be used for testing or for trial purposes.Install methodDocker CLI Object DetectorTensorRT Screenshots of the Frigate UI's System metrics pages
Any other information that may be helpfulNo response |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 11 replies
-
|
your |
Beta Was this translation helpful? Give feedback.
-
|
done, yes now it's saved, but log reads:
and now again blank screen, but gpu reads at 17% instead of zero, so something must have started working
|
Beta Was this translation helpful? Give feedback.
-
|
you need to change the environment variable YOLO_MODELS so it knows to download the yolov7-320 model. in the docker compose |
Beta Was this translation helpful? Give feedback.
-
|
trying to do it in Truenas Scale, but find it bit odd, also tried to add env var to the docker installation but didn't do... |
Beta Was this translation helpful? Give feedback.
-
|
the name is YOLO_MODELS and the value is after = |
Beta Was this translation helpful? Give feedback.
-
|
it is slowing deploying (taking ages) dunno if it will work log: |
Beta Was this translation helpful? Give feedback.
-
|
I suppose it's working now, am I right? |
Beta Was this translation helpful? Give feedback.
-
|
yes, it is running |
Beta Was this translation helpful? Give feedback.
-
|
grazie, danke, merci, gracias, thank you! would have never figured that out! here's smi too: :-) |
Beta Was this translation helpful? Give feedback.
the name is YOLO_MODELS and the value is after =