The DeepStack CPU version is very intensive and needs a lot of CPU resources. Turn on Save DeepStack analysis details and post a screenshot of the details
IP Cam Talk Custom Community DeepStack Model
- Thread starter MikeLud1
- Start date
Will do. Also, I've seen people recommend to check "limit decoding unless required" but it seems to make the UI slower. What do you think about checking that box?
Attachments
sebastiantombs
Known around here
My typical settings, even when I tried the CPU version, have been a maximum of 20 images and I generally use a 250ms timing. It is camera/view dependent and exactly what I'm trying to detect, but 999 is far too many and will keep the CPU churning far too long. Then, if another camera triggers with similar settings, 999 images, it will only get worse. How many cameras are you using DS with?
I only have 2 - 5MP Amcrest cameras which is why this doesn't make sense to me. I've only had them for a week but, based on what I've read on these forums, I should not be seeing that much utilization. I hear of people having 20-30 cameras and it works just fine with 500ms process times.
When I analyze it like this it always does fine. My CPU still jumps straight to 10% but the times are reasonable.
Attachments
You still did not do the below setting, you are running 3 models. When is you CPU jumping up to 100%? If it is only when you are getting an alert, that is normal for the CPU version of DeepStack
sebastiantombs
Known around here
I see a rise in detection times at night because I use two models, combined and dark. During the day I only use combined. This is a situation where the KISS principle applies. As Mike said use "objects:0,combined" and maybe one other model. Three is pushing things yet again.
I changed that setting a couple of hours ago so that shouldn't be the issue. And yes, it jumps up only when I'm getting an alert. If that's normal then I guess we're good.
Attachments
Was the below before the change or after, because the below shows four models running.
This is a screenshot of a more recent event. I must've picked an alert that was before the change.
I don’t understand how that’s still like that. Is there something wrong with these settings?
"Analyze with Deepstack" doesn't use the "Custom models" setting. You'll need to remove the custom models not in use from the "My models" folder. Once you do that, they won't show up in the analysis. Your settings are okay, you can check the log file to confirm it's only using the combined model.
sebastiantombs
Known around here
Analyze with Deepstack is not the same as normal analysis of a clip, either. You need to look at the .dat file to see what actually happens in each clip. Analysis will show everything Deepstack can possibly identify, as well as what real time analysis is set for.
Okay, cool. Is there a specific location the DAT files are saved? I didn’t see an option to map a custom location. I really appreciate your help on this. Hopefully my 2060 helps out.
sebastiantombs
Known around here
The 2060 will make a night and day difference. I'm using an old GTX970 and getting ~50ms detection times with the combined model. Detection times at night are >250ms with combined and dark.
That's awesome. How many cameras do you have? And where did you get the night model? I only saw the three on the original post.