|
|
Just updated to 2.6.2 docker
After some time log get flooded with:
13:26:18:detect_adapter.py: WARNING ⚠️ NMS time limit 0.550s exceeded
13:26:28:detect_adapter.py: WARNING ⚠️ NMS time limit 0.550s exceeded
13:27:25:detect_adapter.py: WARNING ⚠️ NMS time limit 0.550s exceeded
13:27:30:detect_adapter.py: WARNING ⚠️ NMS time limit 0.550s exceeded
It didnt do that ever before.
How do I fix this?
|
|
|
|
|
What module and hardware?
|
|
|
|
|
Thanks very much for your report. Could you please share your System Info tab from your CodeProject.AI Server dashboard?
Thanks,
Sean Ewington
CodeProject
|
|
|
|
|
You're using Coral, right? USB or PCIe?
This error occurs when you're using the YOLOv5 models in Coral, and the non-max suppression method takes too long to sort through all the bounding boxes identified in an image.
What hardware are you running the docker container on, and if possible, could you post an example of an image that's causing this?
cheers
Chris Maunder
|
|
|
|
|
Sorry, forgot to post system info but then could edit post until approved.
not using coral, afaik.
using cuda12.2.2.6.2, upgraded from 12_2.2.5.6 the day before
with Blue Iris 5.9.0.1
CodeProject.AI
Docs
Forums
The Code
CodeProject.AI Explorer
cpai
Offline
2.6.2
☀️
Your self-contained AI server. Need help? Common solutions or ask a Question. Use in your apps: API guide or add your own AI module. FAQs: Blue Iris, Home Assistant, Agent DVR, Wyze cams.
Server version: 2.6.2
System: Docker (cpai)
Operating System: Linux (Ubuntu 22.04)
CPUs: Intel(R) Xeon(R) Gold 6130 CPU @ 2.10GHz (Intel)
1 CPU x 8 cores. 8 logical processors (x64)
GPU (Primary): (NVIDIA), CUDA: (up to: ), Compute: , cuDNN: 8.9.6
System RAM: 16 GiB
Platform: Linux
BuildConfig: Release
Execution Env: Docker
Runtime Env: Production
Runtimes installed:
.NET runtime: 7.0.17
.NET SDK: Not found
Default Python: 3.10.12
Go: Not found
NodeJS: Not found
Rust: Not found
Video adapter info:
System GPU info:
GPU 3D Usage 0%
GPU RAM Usage 0
Global Environment variables:
CPAI_APPROOTPATH = <root>
CPAI_PORT = 32168
|
|
|
|
|
Which module are you running?
cheers
Chris Maunder
|
|
|
|
|
Seems to be limited to face detection. If I turn that off, log's clear.
|
|
|
|
|
Been running 2.6.2 since it's release with YOLOv8 v1.4.3 and FaceProcessing v1.10.2 as well.
Then I restarted the server that runs the Docker environment. Upon restart it keep telling me:
An item with the same key has already been added. Key: FaceProcessing.
I searched all the AppModules for extra FaceProcessing but nothing...
The only thing that would allow me to start CodeProject was to rename the folder FaceProcessing under
C:\ProgramData\CodeProject\AI\docker\modules
Which I bond with the way I run the docker:
docker run --name cuda12_2-2.6.2 -d -p 32168:32168 --gpus all --mount type=bind,source=C:\ProgramData\CodeProject\AI\docker\data,target=/etc/codeproject/ai --mount type=bind,source=C:\ProgramData\CodeProject\AI\docker\modules,target=/app/modules codeproject/ai-server:cuda12_2-2.6.2
This was a clean install, meaning nothing was in the ProgramData\CodeProject\AI\Docker folder until I installed YOLOv8 and Upgraded FaceProcessing. Those were put there on the install.
|
|
|
|
|
This is an issue where you have a module pre-installed then you update or reinstall and it appears in a different folder. This has been fixed in our next release due next week
cheers
Chris Maunder
|
|
|
|
|
Understanding that the YoloV5 model likes to detect house plants and tooth brushes in my backyard, how do I detect all things? Currently in BlueIris under Camera Settings -> Alert -> AI configuration I have "To confirm" set to the default of "person,car,truck,bus,bicycle,vehicle". But I know there is a lot of other stuff, like dog cat... Toothbrush. Can I set "To confirm" to a wild card?
modified 25-Apr-24 11:58am.
|
|
|
|
|
You can just leave it blank and it will detect all the objects that the model detects
|
|
|
|
|
Is there a listing of all the things the model could detect ?
Pete
|
|
|
|
|
|
i recently installed codeproject-ai and pass my coral usb stick through to the docker container. i've downloaded the coral usb module within the cpai interface, but i'm not sure how to make sure all other installed models are running through my coral stick and not cpu.
is there a way to verify this?
|
|
|
|
|
Running Codeproject AI 2.6.2 with object detection (Coral) 2.2.2. And when I watch in the log I get a lot of the following warnings:
16:27:48:objectdetection_coral_adapter.py: WARNING:root:No work in 60.0 seconds, watchdog shutting down TPUs.
16:27:53:objectdetection_coral_adapter.py: WARNING:root:No work in 60.0 seconds, watchdog shutting down TPUs.
16:27:58:objectdetection_coral_adapter.py: WARNING:root:No work in 60.0 seconds, watchdog shutting down TPUs.
16:28:03:objectdetection_coral_adapter.py: WARNING:root:No work in 60.0 seconds, watchdog shutting down TPUs.
16:28:08:objectdetection_coral_adapter.py: WARNING:root:No work in 60.0 seconds, watchdog shutting down TPUs.
16:28:13:objectdetection_coral_adapter.py: WARNING:root:No work in 60.0 seconds, watchdog shutting down TPUs.
16:28:18:objectdetection_coral_adapter.py: WARNING:root:No work in 60.0 seconds, watchdog shutting down TPUs.
What causes this? and how can I fix this?
|
|
|
|
|
It should be booting itself back up as soon as new work arrives, so it shouldn’t be a problem.
|
|
|
|
|
But detection time seems to be a bit slower now, 15-20ms before vs 25ms now? That is really not that much slower but the constant red in the status window makes it seem like something bad is happening.
|
|
|
|
|
In the Blue Iris when a new trigger arrives I get an "AI: error 500". This also happens since the last update. It looks like it needs to wake-up the coral and then has a timeout error.
|
|
|
|
|
Hi,
The majority of the time, even with clear view & multiple reeds showing the same plate (I send images to the server) it reads as an I when should be the number 1
HW13 PBX reported as HWI PBX
As this all goes on in my python program I could program myself to adjust but would be nice if it was a built in feature, perhaps either a country flag or even a template XX##XXX
I know uk private plates might cause issues but better to have the majority.
previously I used paid version of Plate Recognizer (parkpow) which was extremely accurate & had country settings but always got the I & 1 correct but oddly O & 0 mixed up fairly often.
Thanks I/A
|
|
|
|
|
I am building out a large Blueiris setup with 30X 8K cameras. What specific hardware should I keep in mind for Codeproject AI? RAM, GPU, VRAM?
I am looking at running the YOLOv8x model.
Thanks!
|
|
|
|
|
The RTX 3060 12GB would be a good GPU
|
|
|
|
|
Should I care more about card generation GTX 1XXX, RTX 20XX, 30XX ,40XX? Or VRAM?
|
|
|
|
|
|
An nvidia 3050 6Gb would be preferable to an Nvidia Tesla P40 24GB?
Both cards are at the same price point and very similar compute performance. The 3050 is rated around 50w depending of flavor. The P40 wants around 250w. The 3050 has 6Gb~8Gb of VRAM vs 24Gb in the P40.
|
|
|
|