AI Detection Question

[iikingli]

So I am looking at all these yet anotherthreads on AI addons.

Do is there a List (that I cant seem to find) that has the pros and cons of each?

I am not sure which route to go, AI Tools, On Gaurd, something newer, just looking for general directions on comparatives, not stating one is better than the other, but possible, ease of use, features, functionality.. etc..

Just a thought.

 

[sebastiantombs]

You do know that DeepStack is already integrated into BI. All you need to do is download it, click on the AI tab in the BI config and use it. It will run on a CPU that isn't heavily loaded with good results or install the GPU version, needs a CUDA capable card, and get even better results. You can train your won custom models and use them as well to fine tune it to your particular location.

 

[iikingli]

You do know that DeepStack is already integrated into BI. All you need to do is download it, click on the AI tab in the BI config and use it. It will run on a CPU that isn't heavily loaded with good results or install the GPU version, needs a CUDA capable card, and get even better results. You can train your won custom models and use them as well to fine tune it to your particular location.

Yes - I have it installed, I guess I am trying to understand which of the addons, helps reduce some of the false positives, or what the benefits are over DeepStack.

 

[sebastiantombs]

What kind of false positives are you getting? Are you running the CPU or GPU version? What do detection times look like?

 

[whoami ™]

I've been wondering if DS works as well with BI directly as it is does with the latest version of AITool. I've been holding off on renewing my BI updates sub until all the bugs are worked out.

 

[sebastiantombs]

I can only report what I see on my own system. I've got DS running on about ten cameras and using a GTX970 with the GPU version. Detection is excellent with the exception of night time when headlight bloom makes it hard for DS to detect some vehicles. I only see an occasional false positive per day, but that's an exception not the rule.

 

[whoami ™]

Are you able to run multiple instances of DS from BI for parallel processing with GPU yet?

 

[sebastiantombs]

Never even tried that. DS spawns as many instances of Python as it needs when multiple cameras trigger. The GTX970 has no problem handling that at all. Single trigger detections are in the 50-100ms range and multiple trigger detections are in the 100-400ms range depending on the number being handled. That's more than fast enough for DS not to time out.

 

[iikingli]

Figured out one of my issues, dont think it was with Deepstack... now @sebastiantombs do you know If DS works with Dahua or is that only going to be with AI tools. For IVS rules trigging from external sources, in order to identify whats in the image?

 

[IAmATeaf]

Figured out one of my issues, dont think it was with Deepstack... now @sebastiantombs do you know If DS works with Dahua or is that only going to be with AI tools. For IVS rules trigging from external sources, in order to identify whats in the image?

What do you mean works with Dahua? All BI does is send an image to DS to analyse so in theory will work with any cam?

 

[iikingli]

What do you mean works with Dahua? All BI does is send an image to DS to analyse so in theory will work with any cam?

I am talking about it working with External IVS triggers.

 

[sebastiantombs]

Works with Dahua and external triggers for me. Any trigger that BI sees from a camera that is set for motion detection, BI or ONVIF, produces the same set of images being sent to DS for analysis. The brand has nothing to do with that at all.

 

[iikingli]

Works with Dahua and external triggers for me. Any trigger that BI sees from a camera that is set for motion detection, BI or ONVIF, produces the same set of images being sent to DS for analysis. The brand has nothing to do with that at all.

Cool been having issues, might need to try and run DS off my main machine

 

[wittaj]

Or you could let the AI in the Dahua cams (if yours have them) take care of the AI for Blue Iris and only use DS on the cameras that do not have AI.

Who uses Dahua AI capable cameras? Reliable AI for triggering events? Pro's/con's?

It has come to my attention that certain Dahua cameras come with built in AI that distinguishes between human & cars. I was unaware of this. Well, always saw the AI word used in the reviews but never really thought about it because I thought it was a Dahua NVR integration only. I am currently...

ipcamtalk.com

 

[iikingli]

Or you could let the AI in the Dahua cams (if yours have them) take care of the AI for Blue Iris and only use DS on the cameras that do not have AI.

Who uses Dahua AI capable cameras? Reliable AI for triggering events? Pro's/con's?

It has come to my attention that certain Dahua cameras come with built in AI that distinguishes between human & cars. I was unaware of this. Well, always saw the AI word used in the reviews but never really thought about it because I thought it was a Dahua NVR integration only. I am currently...

ipcamtalk.com

Thats what I was doing, but it just says external alert, no tags. In the end I might have to find another solution for DS. As the machine I am running BI on is fine for CPU, sits around 27% as a VM, but of course no decent GPU, but with Deepstack it takes 1.4 -1.6 Seconds to analyze an image. . I have to figure out if I can pass through Deepstack off my main box. 128 GB of RAM with an RTX 3090, if I can make that work it should be fine.

 

[wittaj]

Yes, it will just show external alert. But if you setup the Dahua AI for just human for example, you know every external alert is a human, you just don;t get the little orange figure in BI lol.

 

[iikingli]

Yes, it will just show external alert. But if you setup the Dahua AI for just human for example, you know every external alert is a human, you just don;t get the little orange figure in BI lol.

True, but that little orange guy intrigues me

 

[sebastiantombs]

Where is the orange man when we need him most?