Have you switched from DeepStack to CodeProject.AI? Poll

[CCTVCam]

I would encourage you to give CodeProject.AI a try. Their one-click Windows install is pretty solid and will run fine on CPU, especially if you have an i7 or higher.

Side note, there is no need for regedits, that's old news and was only required for earlier iterations.

There's still some argument for Coral support simply due to CPU load and speed of processing. I somehow doubt adding maybe 20% loading to a 100W cpu (20W if linear) is as efficient as adding a 2W TPU that processes in a fraction of the time. I had a Quaddro graphics card. I sold it without it ever being used because how can I justify something in my system that uses 40W 24/7 but is many times slower than a 2w dongle?

That is useless to 99.99999 percent of blue iris users. That is the beauty of codeproject, easy to install and there is active support from the developers to the blue iris developer. Codeproject has already indicated that they will likely work on Coral support.

Exactly this. Not everyone can code and whilst BI arguably isn't a consumer level project due to the extra knowledge on networking etc being required, most non consumer level users, still aren't computer programmers or capable of writing code.

 

[fenderman]

Exactly this. Not everyone can code and whilst BI arguably isn't a consumer level project due to the extra knowledge on networking etc being required, most non consumer level users, still aren't computer programmers or capable of writing code.

Blue iris is very much a consumer level product. No more networking knowledge is required than an NVR when using with a poe switch, which a very basic minimum level of knowledge of ip addressing. Moreover once the system is setup the user does not need to know any of this. This make the need for a simple solution like code project even more important. They will certainly provide coral support. Makes no sense not to.

 

[TBurt]

I am creating this poll to help users decide if they should switch from DeepStack to CodeProject.AI.

Please post your likes and dislikes, your posts will help improve CodeProject.AI. Also you can go to their discussing board if you find any bugs or have a feature request.

Been running CodeProject.AI just about from the start of their gpu Cuda support and couldn't be happier. I only have a gtx 1060 3 gb gpu. DS uses a lot more resources especially gpu memory on my setup. I would get DS stopped responding randomly all the time. Not so with CodeProject.AI. I can also load more custom models without getting that not responding errors popping up. Much more stable for me by far. People should try it as you could swap back if you needed to.

 

[DanDenver]

BI arguably isn't a consumer level project due to the extra knowledge on networking etc being required

Not sure if this poster has installed and used NVR as well as BI, however this statement is patently false.
I specifically held back from moving to BI for 4 years. My friends had it at work but they used a lot of words that at the time, I was not familiar with.
However, once I did make the move from NVR to BI I found out that they are simply challenging in each of their own ways.

But to characterize one or the other as requiring ”extra knowledge” is not accurate.

Neither approach allows you to just plug it in and it magically starts working.
Both require fundamental networking knowledge.

 

[greg_mitch]

DS stopped working and the next BI update brought CodeProject more front and center. Made the switch and haven't looked back. I think detection is just as reliable and the software seems to not require me to futz with it at all, which is nice.

 

[MikeR33]

I was pretty happy with the DS setup for a few months, didn't experience issues installing or upgrading it but suddenly it just went downhill in reliability (I hadn't changed anything for weeks) and I couldn't find why, clearing out all those orphaned files made no difference any longer (I had automated scheduled jobs running), continually having complete system failures requiring hard power off, constant timeouts after only a couple of hours when running. At this point I made the switch to CodeProject, no more system crashes whatsoever and not had any timeout issues either so happy with CodeProject and looking forward to V2.

 

[Mast3r0fN0n3]

Just as I figured out how to install, and, almost how to use deepstack, I inaccurately, assumed, code project was the latest and greatest "thing". Obviously I now know that is not the case. I originally purchased a ring a ding doorbell hoping that would notify of packages being delivered during the non pandemic. As the more sophisticated know, myself included now, ring a dings are useless as a camera or a doorbell. Fortunately, by some miracle I found ipcam talk and got real cameras with real notifications. Has it been a bumpy ride, you betcha. But it works, which is what I needed. The investment may well have already paid for itself. As far as code project, my set up is probably, "wrong?" compared to what I read. Having said that, my whole life the phrase "if it aint broke dont fix it" comes to mind. I get my regular triggers, no notifications, and important triggers with the vehicle/person showing, which is what I need most. Haven't figured out why the cats don't trigger notifications, but they do trigger recording. Most importantly, my phone dings like crazy if someone is walking the important areas, (I use the doorbell sound). This is what I am after. I have no idea if this is perfect, I'm sure I could improve, but "if it aint broke dont fix it ". Sorry WATTAJ, no "Hat" picks this post.

 

[snakpak]

One of the DeepStack updates last spring broke functionality and significantly reduced accuracy so I quickly switched to Code Project.

I like Code Project better, but to be honest both of these AI systems are pretty brutal. Code Project claims to detect more animals you'd regularly find in North America but it does a very poor job of it. Deep Stack is even worse and can look for giraffes and elephants but I don't have any of those up here in Canada. I care about people and dangerous animals on my rural property.

My cameras are in a rural mountain setting with a large number of animals roaming around, from black bears and grizzly bears to mountain lions (cougars), moose, deer, porcupines, skunks, etc. I cannot rely on either AI system to accurately detect any of these even with crystal clear images. Code Project regularly thinks my dog is a person and a bear is a deer or worse, thinks its nothing at all. A grizzly bear with cubs can be extremely dangerous and I need an AI that works. I will have to figure out how to better train one of these simplistic AIs to better detect dangerous animals.

These AI systems are far too simplistic for accurate detection of anything beyond vehicles and people. Hopefully they will improve significantly over time.

 

[actran]

@snakpak Whether you use deepstack or CodeProject, the model(s) that you use is key to the desired object type detection and/or accuracy.

What model(s) are you using in CodeProject that you don't feel is giving you reliable/consistent results?

Just in case you didn't know, @MikeLud1 custom models are included with CodeProject OOTB.

And you are correct that, for a given situation, you may need to train your own custom model.

 

[snakpak]

@snakpak Whether you use deepstack or CodeProject, the model(s) that you use is key to the desired object type detection and/or accuracy.

What model(s) are you using in CodeProject that you don't feel is giving you reliable/consistent results?

Just in case you didn't know, @MikeLud1 custom models are included with CodeProject OOTB.

And you are correct that, for a given situation, you may need to train your own custom model.

I'm using MikeLud1's IPcam-combined custom model, performance-wise it's pretty good. It's just poor with accuracy of animal detections.

Is there a good thread or guide on how to train my own custom model for use with Code Project? I want to do this in January/February 2023 and am happy to share the model here when I'm done. I'm going to start with images of real bears collected from my security cameras but I suspect I'll need a lot more images to properly build a custom model.

 

[actran]

@snakpak The custom model training steps are not specific to codeproject. Here is a good starting point Train Custom Data · ultralytics/yolov5 Wiki

Credit goes to @MikeLud1 for sharing this URL previously.

BTW, it helps to have a beefy GPU for training.

 

[DanDenver]

People being required to train their own AI for (hopefully) reasonable accuracy demonstrates to anyone reading this thread the infancy of the AI industry.

I do hope you share the models you create. I wanted to do the same, but it ain't quick or easy.

 

[Mast3r0fN0n3]

I'm pretty sure I saw a giraffe, but the cameras didn't trigger. It was about 12:15 am on 1.1.2023 right after I finished the second bottle of champagne.

 

[dkmcgowan]

I was just looking into this. I run deepstack docked container on a separate Linux box that has an extra GPU in it and then put in the IP and port into Blue Iris and it's been working fine. I only use standard models like car, person, truck, nothing really custom...

I saw that Codeproject AI also has a container, so I can just install that, put it on a different port, and I assume change out the IP and port in BI and just compare them with no other changes really right? The API is the same and seems like Codeproject comes with the standard models I use.

Any other things different in BI like names for car, truck, person?

 

[Holbs]

I ran deepstack with Blue Iris during my security tire slasher event. It worked 85% of the time for me for what I needed it to. I eventually switched to Dahua AI in the cameras that worked 100% of the time. I do have cameras that are non-AI that I may go with Codeproject in due time.

 

[Sybertiger]

I've got DS running on two BI systems and so far reasonably satisfied. I use Dahua AI in the cams to trigger BI then have DS filter through those Dahua AI triggered events. In fact, I rely heavily on Dahua trip wires and intrusion rules to the point where I hardly use any of BI's detection tools as I find Dahua AI is superior to BI.

I have an I7-6700 system with no graphics card and don't know what impact if any CodeProject would have on performance. I'd like to switch to CodeProject as DS has been deprecated from more current releases of BI. I'm using the very last BI version that support DS but I don't want to get left behind. If I could convince myself that switching to CodeProject was easy-peasy with assurances I don't break anything I would. It's probably easier than I think and I'm just making a big deal over nothing. I just don't want to have to go back to DS if CodeProject has issues. Also, I have a DS model I put together for USPS delivery detection so I'd have to rebuild that for CodeProject as I understand.

 

[DanDenver]

Same here.... when is the best time to Migrate from DS to Codeproject.
AI is so new relatively speaking and the landscape continues to change fast.
DS was only king of the hill for about 18 months till it got replaced entirely.

Good to hear that the AI in the cameras works well.

 

[MikeLud1]

I have a DS model I put together for USPS delivery detection so I'd have to rebuild that for CodeProject as I understand.

You do not need to rebuild your model. The only thing is if you want to use Object Detection (YOLOv5 .NET) with your model it need to be converted to ONNX model which takes less than 5 minutes. If you are not sure on how to do it I can do the conversion.

 

[davej]

I am curious what BI versions are necessary in order to consider investigating DS or AI? I'm currently stuck at BI 5.5.9.6.

 

[wittaj]

I am curious what BI versions are necessary in order to consider investigating DS or AI? I'm currently stuck at BI 5.5.9.6.

CodeProject was added at 5.5.8, so every version past that is one more chance that Deepstack will not work and one would have to go to CodeProject if they want to use BI AI.