5.5.8 - June 13, 2022 - Code Project’s SenseAI Version 1 - See V2 here https://ipcamtalk.com/threads/codeproject-ai-version-2-0.68030/

I'm making some more popcorn, waiting to see how this plays out some more.

@tofu Are those actual object detection times or face detection times? Is face detection enabled as well? Just curious.
 
I'm making some more popcorn, waiting to see how this plays out some more.

@tofu Are those actual object detection times or face detection times? Is face detection enabled as well? Just curious.


If i watch the log, the times are around 400-500ms. BI says the average is 550ish... either way, it's badddd

facial detection on or off doesn't change anything.

I'm playing with facial detection now -- disabling it in BI and just having sensai learn faces through the dashboard. I'm checking to see if i can just use the tagged name of the faces in BI despite it being disabled.
 
One more question, what card are you using? Mike is using an RTX3090 which gives him a huge advantage over everything else.
 
  • Like
Reactions: looney2ns
I wonder if that could be part of the slow response times. Maybe Mike can shed some light on that aspect.
 
  • Like
Reactions: looney2ns
Wow doesn't really express it well enough! I thought it would be faster, significantly faster. Maybe there's a few kinks to iron out yet.
 
  • Like
Reactions: looney2ns
I made some progress with the facial recognition --

you CAN train multiple images to one tag. Now you don't need dude_1, dude_2, dude_3 -- You can just have DUDE

1) facial recognition must be turned on in BI
2) NEVER press the faces button in BI - - that seems to delete everything.
3) Train the faces through the sensai dashboard. You can select multiple files at once. You can also go back and add files to it under the same tag in the future.

Would be nice to have a list of the files we already used for a specific face tag, though. Not sure if redundancy matters or it's smart enough to realize we already added X image.
 
Or it's like OpenALPR that is slower and a CPU hog when the GPU version is used.

I hope they get it worked out or that Deepstack will still function in future updates. My DS has been rock solid with quick enough response times that I don't see a need to switch. YMMV
 
Here is mine with no GPU. But in my case this runs perfect on my BI install. The times really mean nothing to me as long as it works great. I also think Code project AI detects way better than DS and more accurate. Of course every install is different, I know.
 

Attachments

  • Screenshot 2022-08-11 161857.png
    Screenshot 2022-08-11 161857.png
    815.9 KB · Views: 71
As an aside, CPU and GPU models, if a GPU is being used, would be a little more informative when posting benchmarks.


The slow times is an issue with the server logging the modules are fast see below

Can you elaborate on that? I don't understand what you're saying here. I would think that using the custom model, which is considerably smaller, it would speed things up a bit more for both CPU and GPU operations.
 
  • Like
Reactions: looney2ns
As an aside, CPU and GPU models, if a GPU is being used, would be a little more informative when posting benchmarks.




Can you elaborate on that? I don't understand what you're saying here. I would think that using the custom model, which is considerably smaller, it would speed things up a bit more for both CPU and GPU operations.
When they first added custom model support I was see closed to GPU speeds using a CPU. One of the versions before GPU support they added enhanced logging and that is when the slower detection times started. When using the GUI it looks like it is bypassing the some part of the logging and the modules run faster. Below is a post that notes they think they found the issue.

1660264783539.png
 
Thanks Mike.
 
So just confirming, what is everyone using for BI --> Code project in terms of models?
I'm using custom models for all my cameras. Most of the cameras are using ipcam-general however, there are a few that are using ipcam-dark at night. I also added a USPS custom model that I'm using on a cloned camera to trigger mail delivery.