IP Cam Talk Custom Community DeepStack Model

jbaliza

n3wb
Joined
Aug 8, 2022
Messages
1
Reaction score
0
How should one configure BlueIris to call the "general" custom model when deepstack is run from Docker/Jetson Nano?

I tried (in the camera's "Trigger/Artificial Intelligence" form):
  • "objects:0,general" or "objects:0,custom/general": deepstack does not get called (confirmed with wireshark)
  • "general" : default detection gets called (/v1/vision/detection)
The general model runs fine when I call it directly (using /v1/vision/custom/general instead of /v1/vision/detection), so there is no problem on the deepstack side.

This how I load deepstack:
I put general.pt in ~/aimodels/
and started deepstack with:
sudo docker run -d --log-driver syslog --runtime nvidia --name deepstack --restart unless-stopped -e VISION-DETECTION=True -e MODE=High -v /home/myuser/aimodels:/modelstore/detection -p 80:5000 deepquestai/deepstack:jetpack-2021.09.1

I cannot run the combined model, as I found that loading that model provokes an out of memory on my 4GB Jetson Nano.

BI: 5.5.3.7 on Win10
Hi friend, did you find an answer? I'm having the same problem here.

Running on docker, if I call the general api directly it works.
1660848012279.png
 

jairbj

n3wb
Joined
Aug 9, 2022
Messages
6
Reaction score
0
Location
Brazil
Hi friend, did you find an answer? I'm having the same problem here.

Running on docker, if I call the general api directly it works.
View attachment 137065
I had the same problem and what solved to me was point Blueiris to a folder with the models, even if Deepstack is grabbing it from another directory and is running on a docker container.

1660851976937.png

In this case Deepstack is not looking into this folder. I just copied it into this directory then Blueiris "knows" this model exists.
 

Freddan101

n3wb
Joined
Apr 4, 2016
Messages
26
Reaction score
24
The person that gave me some more deer images made a deer only model, this model was trained with about 600 images. If anyone wants to try it out see attached.
I've been trying out the deer.pt, and while it seems to detect deer it works a little bit too well... Now pretty much everything is identified as "deer". Don't think there is any way to fine-tune this without creating an own model, right? Anybody else experiencing this with deer.pt?

1661090716212.png
1661090459640.png
1661090802093.png
 

MikeLud1

IPCT Contributor
Joined
Apr 5, 2017
Messages
2,141
Reaction score
4,118
Location
Brooklyn, NY
I see on @MikeLud1's github there are two custom model projects DeepStack-Security-Camera-Models and CodeProject.AI-Custom-IPcam-Models. It appears like the CodeProject.AI-Custom-IPcam-Models has been updated more recently? I'm currently using DS not CP, are the CP pt models backwards compatible with DS? Are they even any different/newer/better/etc?

Thanks
They are the same models just different file names. I am going to be updating them before the end of this year
 

EvanVanVan

Pulling my weight
Joined
Jul 29, 2022
Messages
135
Reaction score
102
Location
NJ
They are the same models just different file names. I am going to be updating them before the end of this year
Nice, thank you for the reply!

As a quick suggestion, the one animal I was hoping for that's missing from the animal models, would be a moose. Any plans on adding it? (Although I suspect there's a good chance it would be flagged as a Horse at a minimum lol.)
 

Lucas Tam

n3wb
Joined
Apr 12, 2017
Messages
27
Reaction score
5
Hi all,

I'm using CodeProject Sense AI w/ ipcam-general model, and I noticed that this model detects cars, trucks, buses, etc. with the "vehicle" label? With the IPCam-Combined model, it detects the individual type of vehicle.

I thought these models were built from MIke Lud's models - but on Mike's GitHub page IPCam-General doesn't have a vehicle label - does anyone know why Sense AI's model would return vehicle for IPCam-General? Were these models changed for Sense AI?

UPDATE ... just saw this post:

I see on @MikeLud1's github there are two custom model projects DeepStack-Security-Camera-Models and CodeProject.AI-Custom-IPcam-Models. It appears like the CodeProject.AI-Custom-IPcam-Models has been updated more recently? I'm currently using DS not CP, are the CP pt models backwards compatible with DS? Are they even any different/newer/better/etc?
This examples why the labels are different - the Sense AI ones have different labels than the older DS models!

@MikeLud1 - For Vehicle's what's grouped under that label - did you include motorcycles and bicycles too?

Also, I think the simplification is great fro general CCTV AI usage.
 
Last edited:

sebastiantombs

Known around here
Joined
Dec 28, 2019
Messages
11,511
Reaction score
27,690
Location
New Jersey
Mike simplified the labels in the "general" model after a discussion here on IPCT to "person or vehicle". The combined model was left with the more inclusive list as part of that discussion, if I remember correctly. The AI, DeepSatck or SenseAI, can only identify based on the object names in the model and labels them accordingly.
 

Lucas Tam

n3wb
Joined
Apr 12, 2017
Messages
27
Reaction score
5
I noticed the black squirrels seem to trigger Person Detection quite regularly with high confidence scores with the IPCAM General model, anyone else have this issue?

ipcam-general returns a person (BI reports 77%)
ipcam-animal returns nothing
ipcam-dark returns nothing
ipcam-combnined returns nothing
(human) ActionNetv2 returns nothing

Anyone know how to export the SenseAI detection images so that I can provide some samples?
 
Last edited:

rdxny

Getting the hang of it
Joined
Oct 15, 2022
Messages
139
Reaction score
81
Location
ny
I'm using CodeProject Sense AI default. One camera detects a car entering driveway fine, another does not trigger an alert despite saying 86%... so before II post all my settings, my question is : Can the fact that the one camera triggered an alert stop the other from triggering at essentially the same time? If they should both trigger, then I have a different issue and can post screen grabs of settings if that helps, Thanks.
 

rdxny

Getting the hang of it
Joined
Oct 15, 2022
Messages
139
Reaction score
81
Location
ny
Change the AI Server port in BI from 5000 to 32168 and see if that helps
Trying that. We will see tomorrow. If no help, maybe go back to Yolo vs ,Net? or a custom model? I will update with results, thanks.
 

rdxny

Getting the hang of it
Joined
Oct 15, 2022
Messages
139
Reaction score
81
Location
ny
I'm trying yolov5l... mixed results, Can you confirm the settings I highlighted? I suspect one or more is incorrect (thinking use custom folder should be on... and unsure re the object detection settings) Thanks!

main.pngmotiom sensor.pngai.png
 

Vettester

Getting comfortable
Joined
Feb 5, 2017
Messages
740
Reaction score
693
Can you confirm the settings I highlighted?
In your first screenshot check the "use custom model folder" option. Make sure your custom model folder contains the yolov5I.pt file.
In the second screenshot enable object detection (you can leave the settings within the popup window at the default values).
The last screenshot looks OK but you might want to adjust the min confidence level.
 
Top