CodeProject.AI Version 2.0

MikeLud1 said:
ipcam-general detected the person. This model will work better then ipcam-combined
Click to expand...
Mike: Can you elaborate on how the "ipcam-general" works better that "IPcam-combined"? My nemeses is deer, and I think "deer" is contained in "combined" but not in "general"? ( I tried "Critters" but had many many false alerts... bugs, moths, etc that get ignored with "combined")
If General is better in substantial ways, can "Deer" be added to general? :) Thanks...
 
  • Like
Reactions: David L
rdxny said:
Mike: Can you elaborate on how the "ipcam-general" works better that "IPcam-combined"? My nemeses is deer, and I think "deer" is contained in "combined" but not in "general"? ( I tried "Critters" but had many many false alerts... bugs, moths, etc that get ignored with "combined")
If General is better in substantial ways, can "Deer" be added to general? :) Thanks...
Click to expand...
I second this, we just moved and have white-tailed deer all over this area. Had a family of about 8 the other day in our backyard, can't wait to get my CAMs back up...
 
rdxny said:
Mike: Can you elaborate on how the "ipcam-general" works better that "IPcam-combined"? My nemeses is deer, and I think "deer" is contained in "combined" but not in "general"? ( I tried "Critters" but had many many false alerts... bugs, moths, etc that get ignored with "combined")
If General is better in substantial ways, can "Deer" be added to general? :) Thanks...
Click to expand...
The reason General work better for person detection is because it only has two objects, person and vehicle. The more object added to the model the accuracy will decrease because the model is a fixed size
 
  • Like
Reactions: bug99, rdxny and David L
MikeLud1 said:
The reason General work better for person detection is because it only has two objects, person and vehicle. The more object added to the model the accuracy will decrease because the model is a fixed size
Click to expand...

I thought i also read that General had "dark models"? Are they also in Combined?

What does "objects:0" do in the Custom models filed if IPcam-general is used?
 
bug99 said:
I thought i also read that General had "dark models"? Are they also in Combined?
Click to expand...
The General model was trained with adding the dark model images. The Combined model was not trained with any dark model images
bug99 said:
What does "objects:0" do in the Custom models filed if IPcam-general is used?
Click to expand...
It disable the default object detection.
 
  • Like
Reactions: bug99 and David L
David L said:
Please don't take this the wrong way but this is a good thread to read that will help with your blurring...you may want to consider adjusting your shutter speed...

The Importance of Focal Length over MP in camera selection

At the urging of several folks here, I created a thread to show the importance of focal length and how focal length can be more important than megapixels (MP). I mentioned some of this in the post regarding The Hookup’s latest video demonstrating different cameras, including one sold from a...
ipcamtalk.com ipcamtalk.com
Click to expand...
No worries. that is a great post. I am glad someone put that out.

I, like the poster (@wittaj) started a long time ago on the wrong foot camera wise. This one was the first, and only 4MP (not low light) camera. I even think it is 2.4mm vs 3.6mm. It is hard to get far enough away for a front door camera to get the right field of view and capture people at and approaching the door. I have 4 cameras in and around the front of the house. In my opinion, this should be replaced, but is adequate for the 4ft to 15 ft distance from the camera to the furthest that i care about, and i leave it in color mode 24-7. To my eyes, the image of a slow moving person at the door is easily identifiable during the day and recognizable at night. I don't quite get why CPAI feels that it is not even detectable as human (person). Is it the blur?

If i increase the shutter speed above 1/50, the image is too dark to use at night.
 
Last edited:
  • Like
Reactions: David L
bug99 said:
No worries. that is a great post. I am glad someone put that out.

I, like the poster (@wittaj) stated a long time ago on the wrong foot camera wise. This one was the first, and only 4MP (not low light) camera. I even think it is 2.4mm vs 3.6mm. It is hard to get far enough away for a front door camera to get the right field of view and capture people at and approaching the door. I have 4 cameras in and around the front of the house. In my opinion, this should be replaced, but is adequate for the 4ft to 15 ft distance from the camera to the furthest that i care about, and i leave it in color mode 24-7. To my eyes, the image of a slow moving person at the door is easily identifiable during the day and recognizable at night. I don't quite get why CPAI feels that it is not even detectable as human (person). Is it the blur?

If i increase the shutter speed above 1/50, the image is too dark to use at night.
Click to expand...
A fish-eye lens CAM may work. I had a Doorbell CAM that was pretty good at closer views. Also used it for CPAI. Can't answer the blur question but we know as AI gets better, we will need better images, that is to detect hats, glasses, objects, etc.
 
I am tying to understand how the CPAI determines and processes motion/static objects to better understand why detections and IDs are going poorly on my end. As I understand it, motion is detected by BI (could be from the camera, but I don’t think so in my setup) and then sends the initial trigger alert (one and only one frame) to the AI to process against a database of models. The AI is basically a filter for the BI motion detections + identification. If this is basically correct, how does the AI know about static objects? Does it capture an early frame (say a few minutes earlier with no motion triggers) and compare the new frame against it for % change in pixels?
 
tofu said:
Yes, it defaults to medium, but I manually set it to small. Medium results in very high latency, worse than even an old p400 video card, so it defeats the purpose.

If the coral can't reliably be used at "small," there isn't much purpose for it.

The alternative is that the default trained models are just very poor compared to yolo -- resulting in inaccuracies for cctv
Click to expand...

You don't have to use small. I understand Medium is pretty much the default most use and although I use medium, I feed it 1080P at 4,000kbs because I also record BVR so need a decent sub stream quality or there's no point.

Even on CPU only with a Yolo 6 (whcih isn't recommended for CPU only particularly), I record detections of around 250ms on average. Not the fastest but neither an issue. My profile is to look for person, bicycle, truck, car.



Daniel15 said:
I've got the dual-edge one:
coral.ai

M.2 Accelerator with Dual Edge TPU | Coral

Integrate two Edge TPUs into legacy and new systems using an M.2 (E key) interface.
coral.ai coral.ai

I use it with this PCIe adapter:
www.makerfabs.com

Dual Edge TPU Adapter - PCIe x1 Low Profile

Dual Edge TPU Adapter is designed for Coral m.2 Accelerator with Dual Edge TPU to be used on a system with PCIe x1 slot available.
www.makerfabs.com www.makerfabs.com

Works great. The two TPUs are exposed as two separate PCIe devices, so you can use them in two different VMs if you wanted to.
Click to expand...

Thanks Daniel. Could you just explain, are you runnign Windows and what additional software / firmware do you need to run to make it work?
 
Lurked and learned here for many years, but this is my first post. Trying to setup CodeProject AI Server, but I cannot start the service. I had V2.1.9 and have just tried V2.1.10 both have the same issue.
I have googled, read the forum and there is nothing I have not tried ( except the one thing I need: the solution). Turned off firewall, antivirus, pihole, and all windows security features, uninstallled, restarted, reinstalled, cleaned all references from registry, same result. Has anyone got/found a solution to this? I have put way too much time into this and this will be my last attempt before giving up on CodeProject AI Server. I was able to install and run it on another windows machine.
 

Attachments

  • Screenshot 2023-08-03 211503.png
    Screenshot 2023-08-03 211503.png
    8.1 KB · Views: 28
Last edited:
MikeLud1 said:
The General model was trained with adding the dark model images. The Combined model was not trained with any dark model images

It disable the default object detection.
Click to expand...

Mike, Apparently "objects:0" disables vehicle detection, as the cameras that i added that to now drop them after detection. This is weird. How is a car any more of an object than a person? I assumed that you were referring to the "Default object detection" selection in the AI main tab (which i think is a model choice). [EDIT: apparently I was wrong about this. apparently the issue is that vehicle is not a superset of cars, etc. The "to confirm" of "vehicle" needs to be added to the list "to confirm". On one hand that seems obvious, on the other, it seems like it should be a superset of cars, trucks, etc.]

Does the dark model images used in General add all the dark objects back in, or is it a subset with only [dark] person and vehicle?
 
Last edited:
MikeLud1 said:
When I added the dark images to the General model it merged the objects to person and vehicle.
Click to expand...
Thanks for clearing that up. I hadn't realized that you built the models yourself. Nice! On a somewhat related note:

What is meant by large vs small model size in CPAI Object detection? Are you referring to the number of images in the data set, or the size of each image in the data set, or both?
 
Last edited:
MikeLud1 said:
Version 2.1.10 was released yesterday, there are some minor bugs and should be fixed later today.
Click to expand...
Hey Mike, is this one of the minor bugs? I've been running this since yesterday and everything seems to be working, but the number of detections processed is not being reported.

Screen Shot 2023-08-04 at 5.46.16 PM.png
 
You must log in or register to reply here.
Top Bottom