5.5.8 - June 13, 2022 - Code Project’s SenseAI,

Village Guy

Pulling my weight
Joined
May 6, 2020
Messages
255
Reaction score
117
Location
UK
My P400 has 2GB of RAM.
I'm guessing you were running object detection net and not Python modules?
Good to know 2GB works albeit only partially at this time.
 
Last edited:

kklee

Pulling my weight
Joined
May 9, 2020
Messages
182
Reaction score
194
Location
Vancouver, BC
According to the console, it's using custom object detection and not Python, which I guess is the default since the only configuration change I made was to disable face and scene detection. I'm still learning the ins and outs of SenseAI and fine tuning it for my needs.
 

Village Guy

Pulling my weight
Joined
May 6, 2020
Messages
255
Reaction score
117
Location
UK
According to the console, it's using custom object detection and not Python, which I guess is the default since the only configuration change I made was to disable face and scene detection. I'm still learning the ins and outs of SenseAI and fine tuning it for my needs.
Believe me your not alone, I'm a novice picking up tidbits from the more we'll informed ;)
 

toastie

Getting comfortable
Joined
Sep 30, 2018
Messages
243
Reaction score
72
Location
UK
I apologize, prompted by another thread on IPCT (Videos all black beyond 5 days) I looked in my BI Global>settings>Clips and archiving to see the Folder AI127input which I'd not noticed before, so it hasn't been configured, nor do I know what I would need to enter, or anything.
In the thread, member pbc had Folders "aiinput" and "LPR", I'm not interested in LPR however, now I am left wondering if my un-configured Folder AI127input is the cause of Code Project's AI not working here, a long shot I must admit.
What's the info about the Folders labelled aiinput or AI127input"?
Here apart from the C: Drive, 100% of videos now go to the D: Drive, 99% on New, Alerts 2G, none to Stored, none to AI127input.
 

Village Guy

Pulling my weight
Joined
May 6, 2020
Messages
255
Reaction score
117
Location
UK
I apologize, prompted by another thread on IPCT (Videos all black beyond 5 days) I looked in my BI Global>settings>Clips and archiving to see the Folder AI127input which I'd not noticed before, so it hasn't been configured, nor do I know what I would need to enter, or anything.
In the thread, member pbc had Folders "aiinput" and "LPR", I'm not interested in LPR however, now I am left wondering if my un-configured Folder AI127input is the cause of Code Project's AI not working here, a long shot I must admit.
What's the info about the Folders labelled aiinput or AI127input"?
Here apart from the C: Drive, 100% of videos now go to the D: Drive, 99% on New, Alerts 2G, none to Stored, none to AI127input.
Initially I would forget about folders and simply test CodeProjectAI using the built-in test feature. If that works your good to setup working folders.
 

actran

Pulling my weight
Joined
May 8, 2016
Messages
409
Reaction score
208
BI v5.6.0.8, CodeProject.AI v1.5.6-Beta_0002 CPU only, i7-7700 CPU. See attached files.

I have 1 cloned camera dedicated to delivery.pt (v1.4) custom model only. It doesn't recognize USPS, and I see that also when I inspect the dat file.

If I test a snapshot manually in Explorer UI, it finds USPS.

This configuration did work earlier today with Amazon truck (see another screenshot).

Why didn't it work for USPS in real-action (versus Explorer test)? Anything else I can check/debug?

I have the dat file but I'm not sure I can upload it here as an attachment.

P.S. As a side note, I have success with other custom models like ipcam-general.pt on 7+ cams. And CodeProject.AI has been easy to install and use with BI, and rock solid so far.
 

Attachments

Last edited:

MikeLud1

IPCT Contributor
Joined
Apr 5, 2017
Messages
1,155
Reaction score
2,567
Location
Brooklyn, NY
BI v5.6.0.8, CodeProject.AI v1.5.6-Beta_0002 CPU only, i7-7700 CPU. See attached files.

I have 1 cloned camera dedicated to delivery.pt (v1.4) custom model only. It doesn't recognize USPS, and I see that also when I inspect the dat file.

If I test a snapshot manually in Explorer UI, it finds USPS.

This configuration did work earlier today with Amazon truck (see another screenshot).

Why didn't it work for USPS in real-action (versus Explorer test)? Anything else I can check/debug?

I have the dat file but I'm not sure I can upload it here as an attachment.

P.S. As a side note, I have success with other custom models like ipcam-general.pt on 7+ cams. And CodeProject.AI has been easy to install and use with BI, and rock solid so far.
It could be you are overloading your CPU by setting analyze one each to 50ms, try 250ms of 500ms. Also you do not need to use mainstream.
 

VideoDad

Getting the hang of it
Joined
Apr 13, 2022
Messages
58
Reaction score
78
Location
USA
It could be you are overloading your CPU by setting analyze one each to 50ms, try 250ms of 500ms. Also you do not need to use mainstream.
I agree, you need to give AI more time to analyze the video and find the logo. I have mine set to 5 images,1 second apart. That gives time for the full truck and logo to get into view. 50ms is a little too aggressive, even if you have a GPU, but if you have CPU give it even more headroom.
 
Last edited:

CrazyAsYou

Pulling my weight
Joined
Mar 28, 2018
Messages
171
Reaction score
183
Location
England, Near Sheffield
Working great on a GTX970 with 4GB - I'm only running custom with ipcam-general - Shame I still can't get it running in GPU mode with a GTX1650 (also 4GB ram)
1662305757695.png

1662305638355.png
 

actran

Pulling my weight
Joined
May 8, 2016
Messages
409
Reaction score
208
I'm curious if anyone has come across a buyer's guide on Nvidia cards for AI use case (I'm not interested in cards for gaming at all, in fact, are there cards that doesn't have any gaming bloat like display ports).

I was hoping to understand what card to buy based on price, AI performance,GPU RAM, compatibility with AI software and power consumption.

A quality card with decent performance at low low power consumption is the ideal card for me, since I will be leaving BI PC on 24x7.

Related, I assume the card you choose will be different if you want one for BI AI prediction workload versus for custom model training.
 
Last edited:
Joined
Dec 28, 2019
Messages
11,503
Reaction score
27,467
Location
New Jersey
NVidia is running a big sale this weekend. The 1060 and the 2060 are in the $200 + - range but band new in the box. Either one will work well and I think the 2060 would work the best. If I remember the ad correctly, ~$249. Check NewEgg and some others.

Edit - The EVGA site is showing the RTX2060 SC, 6GB, for $229.

Edit, again - Check antonline.com The RTX2060 SC is about ten bucks more than a T600. Great prices on other NVidia cards as well, but the sale is over tomorrow.
 
Last edited:

toastie

Getting comfortable
Joined
Sep 30, 2018
Messages
243
Reaction score
72
Location
UK
It's encouraging to read reports from those getting Code Project working, here successes are only partial.
In the Dashboard Object Detection works at Detecting Scenes, downtown, home office, classroom etc.
Using Custom Object Detection nothing is found with ipcam-combined or any other choice, "No predictions returned", API server is online apparently.

Looking on the Nvidia website my Nvidia Quadra T600 Turing comes out as a creditable performer with a Compute Capability rating of 7.5 with its 640 CUDA parallel processing cores.
Yesterday I deleted 1.5.6-Beta0002 and re-installed.
AI is apparently still not fully working. I'm on CUDA 11.7 and used the install script for cuDNN.

Despite having uninstalled DeepStack at least a week ago there remained remnants of Deepstack previous presence on my PC. I understand that I shouldn't need to edit the Reg to use Code Project AI's custom object detection assets, because it will been changed with the install of the latest version of Code Project AI. I checked anyway and and it looks OK, but it seems strange that I'm still seeing Deepstack entries in the Reg after I un-installed DeepStack.
What would happen if someone installed Code Project AI to work with Blue Iris who had never installed DeepStack?

If in BI Global settings>AI I open a browser to Code Project AI's Dashboard, the Service API Url is yet if I open the Code Project AI Dashboard from W10 the Url in the window is I though these would be the same but apparently not.
The way Code Project AI contacts the server is different in BI than using W10 Programs to access the Dashboard with a browser. Is this to be expected?
 

IAmATeaf

Known around here
Joined
Jan 13, 2019
Messages
2,780
Reaction score
2,344
Location
United Kingdom
Tighter integration is coming from what I’ve read just not there at the mo.

localhost is the same as 127.0.0.1 but as @MikeLud1 has pointed out you need the port number in order to both access and connect.
 

toastie

Getting comfortable
Joined
Sep 30, 2018
Messages
243
Reaction score
72
Location
UK
The number of the Port 5000 is already included in it's own box to the right of the box entitled Use Ai Server on IP/port. If I add the port twice, as in 127.0.0.1:5000 and another 5000 in the port box no better results.
I realized at the time that my post could lead to this mis-conclusion as IPCT did its own thing with my URLs and I didn't know how to edit my post to avoid any confusion.
Anyway either way of adding 5000 or both didn't help getting AI working on my alert clips, just a lot of red circles with white crosses.
 

JL-F1

Getting the hang of it
Joined
Jun 12, 2020
Messages
81
Reaction score
55
Location
USA
I'm curious if anyone has come across a buyer's guide on Nvidia cards for AI use case (I'm not interested in cards for gaming at all, in fact, are there cards that doesn't have any gaming bloat like display ports).

I was hoping to understand what card to buy based on price, AI performance,GPU RAM, compatibility with AI software and power consumption.

A quality card with decent performance at low low power consumption is the ideal card for me, since I will be leaving BI PC on 24x7.

Related, I assume the card you choose will be different if you want one for BI AI prediction workload versus for custom model training.
Also curious about this, for example buy a 3060 with 12GB ram or a 3060ti with 6GB

Ram amount seeems very big with Code Sense
 

woolfman72

Getting the hang of it
Joined
Dec 28, 2015
Messages
139
Reaction score
66
Also curious about this, for example buy a 3060 with 12GB ram or a 3060ti with 6GB

Ram amount seeems very big with Code Sense
going to be starting over as I need to replace a failing hard drive. currently using CPU DS, going to switch to senseAI but I was wondering the same. I have choices on cards. 2070s, 2080ti or a 3060ti is ram amount more important?
 
Joined
May 27, 2020
Messages
5
Reaction score
0
Location
Canada
I'm trying to run the GPU version in docker on Debian. My GPU is a Quadro K620.
The modules load but I get an error that CUDA is out of memory. When I test images via the CP.AI Explorer page, it seems to work, but the AI-Tool just hangs forever. How can I get AI-Tool to work with the GPU version?
How do I load only the modules required to detect people/animals/cars? Will that save GPU memory and allow it to run properly?

+-----------------------------------------------------------------------------+
| NVIDIA-SMI 515.65.01 Driver Version: 515.65.01 CUDA Version: 11.7 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Quadro K620 Off | 00000000:01:00.0 Off | N/A |
| 34% 45C P0 1W / 30W | 0MiB / 2048MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+
 
Top