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/

Revisiting this question:

Anyone know if CodeProject plans to be compatible with Jetson Nano?

I took a shot at it tonight and got an error related to deploying an amd64 image on an arm64 platform.
 
Is anyone finding the AI reliable? I am really struggling to get sense out of it. I mean, in what world could 2 large pot plants be mistaken for a cat? At 64% certainty?

junk.png

And, when my cat DOES do a leisurely stroll across the room, at least 50% of the time, he gets ignored (although aften the AI will track him (motion rectangles), but then just discard the alert and say nothing found). Upload below shows my large cat walking at a leisurely pace across the room but not being tagged as anything of interest.

junk1.png

Uploads below show cat being recognised as a cat, then as a bird.

junk2.png

junk0.png

The fact that the cats do get picked up shows the AI is working, but the fact that they are often tagged as people or objects (toilet?), or ignored completely, and other random objects get tagged as cats, leaves me with something I simply can't rely on regarding triggers. (This morning, husband and cat in the kitchen. In spite of the huge size difference between them, they both got tagged as cats? Bizarre. Is the only way to get a reliable AI to do a DIY?
 
Have you tried using the yolov5l model? It’s not 100% accurate, but I have found that it works well detecting our dog.

No - I was of the understanding the ipcam models are the same, but updated versions, of the yolov5l? I will give it a try though, can't hurt.
 
No - I was of the understanding the ipcam models are the same, but updated versions, of the yolov5l?
They are not the same. I don't have a cat but there was one in my driveway this morning so I ran a capture through AI explorer. The yolo model was 83% confident that it was a dog while the ipcam-animal model was 69% confident that it was a fox and ipcam-combined model was 46% confident that it was a fox.

Screen Shot 2022-11-10 at 8.36.32 AM.png

Screen Shot 2022-11-10 at 8.34.47 AM.png

Screen Shot 2022-11-10 at 8.33.57 AM.png
 
Maybe there should be a model similar to ipcam-general, where everything that's not a person or a vehicle is an animal. In my situation, I'm perfectly fine with a raccoon, possum, squirrel, pig, bird, fox, horse, cow, cat or dog having a generic "animal" label.
 
Maybe there should be a model similar to ipcam-general, where everything that's not a person or a vehicle is an animal. In my situation, I'm perfectly fine with a raccoon, possum, squirrel, pig, bird, fox, horse, cow, cat or dog having a generic "animal" label.
That is my plan for ipcam-combined "person, vehicle, animal". Once I am finished with the APLR module I am going start retraining my models.
 
Really looking forward to Coral TPU support!


I'm looking forward to this as well. Supposed to be much faster than a CUDA graphics card for a fraction of the price and only a 5w energy draw. The big issue atm is they're sold out and the only people selling them are scalpers. I think they should be $59 but most are approaching $200 on sites such as ebay. Hopefully we'll get an easy 1 click installer / configurator and support soon for Codeproject and they'll be back in stock.
 
That is my plan for ipcam-combined "person, vehicle, animal". Once I am finished with the APLR module I am going start retraining my models.

Yep unless looking for a specific animal by using a camera for animal spotting, for cctv purposes I think person, animal, car, bike, truck, bus, will suffice. An ability to add back in extras to create a custom model would be handy for those wanting as cut down version but with non run of the mill requirements eg someone spotting coyotes, or using it at an airport (plane) or at a marina (boat). For most though, I think PACBTB will suffice. I could see someone wanting knfie or gun, but usually they'd be apparent from the video and it would trigger for person if that were the case in any event.
 
Last edited:
  • Like
Reactions: sebastiantombs
I’m having trouble with CodeProjectAI sending me “nothing found” like leaves, rain, shadow, etc notifications. I only want notifications if there’s a person, car, etc. DeepStack seemed to work just perfect. Somehow back in September I must have upgraded to CPAI. How can I fix CPAI or go back to DeepStack? Thank’s.
 
I'm looking forward to this as well. Supposed to be much faster than a CUDA graphics card for a fraction of the price and only a 5w energy draw. The big issue atm is they're sold out and the only people selling them are scalpers. I think they should be $59 but most are approaching $200 on sites such as ebay. Hopefully we'll get an easy 1 click installer / configurator and support soon for Codeproject and they'll be back in stock.

I was lucky enough to get my USB Coral stick back when they were $59. Now just waiting for a reason to use it :-\
 
CPAI works great so I'm wondering if you have something mis-configured. Post a screenshot of the BI settings AI tab and the AI tab from the trigger tab of one of your cameras.
Okay. As soon as I get back out to were BI computer is I will get screen shots. Thanks. But to be honest, I think it’s working but it sends “nothing found“ alerts all day. And I don’t want the “nothing found” push notifications. when it’s cloudy, “nothing found” alerts are almost gone. When everything was working beautifully i didn’t get alerts like cat’s, dogs, birds, leaves, shadows, rain, raccoons, etc. It was usually people and cars only. that’s what I’m trying to get back working.
 
It was usually people and cars only. that’s what I’m trying to get back working.
It should be an easy fix. Here's a screenshot of the settings for one of my cameras that you can use for a reference. I'm not going to post my general AI settings because I'm running CP AI on multiple computers and don't want to cause any confusion.

Screen Shot 2022-11-14 at 6.08.30 PM.png
 
Last edited:
So im confused on the facial rec part. I have 1 image uploaded to BI in faces and the same image is registered in AI. If i look directly at the camera, it recognizes me about 70% of the time. But if its from the side or a different angle than the original picture, it just says person. Do i need to upload many different pictures to the BI faces and and to the AI? How would I label them? like person1 person2, etc. in BI and can you have multiple different pictures under the same name in AI? Also, Im using the ipcam-general model, is there another one i should be using?
 
So im confused on the facial rec part. I have 1 image uploaded to BI in faces and the same image is registered in AI. If i look directly at the camera, it recognizes me about 70% of the time. But if its from the side or a different angle than the original picture, it just says person. Do i need to upload many different pictures to the BI faces and and to the AI? How would I label them? like person1 person2, etc. in BI and can you have multiple different pictures under the same name in AI? Also, Im using the ipcam-general model, is there another one i should be using?

@swilliams76360 ipcam-general model has nothing to do with face detection. It has only 2 labels: vehicle and person.

If you just want to face detection, then don't use any of the ipcam-* models. In other words, these ipcam models have nothing to do with face detection.

To understand what is happening with each triggered event, use the BI status UI as shown below. The screenshot below for example shows the results of ipcam-general. Face detection should be separate line item if used.

AI details.png