5.4.9 - June 25, 2021
- Thread starter fenderman
- Start date
IAmATeaf
Known around here
Just tried .6 for me i still get no alerts being flagged but the clip is being recorded which I hadn’t noticed or spotted before.
Would the changes made for DS have an impact on alerts? Maybe I need to take another look at the settings, wondering if there’s something that needs to be changed but with 5.4.8.2 I do get the trigger event being flagged in the timeline , with any of the 5.4.9.x I don’t get anything being flagged.
Would the changes made for DS have an impact on alerts? Maybe I need to take another look at the settings, wondering if there’s something that needs to be changed but with 5.4.8.2 I do get the trigger event being flagged in the timeline , with any of the 5.4.9.x I don’t get anything being flagged.
sebastiantombs
Known around here
I emailed BI support with the results I've had switching from CPU to GPU versions. I'm waiting to see if I get a response and what Ken has to say about it.
@IAmATeaf What do your detection times look like in 5.4.8.2 and, if you made note of them, in 5.4.9.x? I think that once that time goes past six or seven hundred mlliseconds DS times out and detections won't happen.
@IAmATeaf What do your detection times look like in 5.4.8.2 and, if you made note of them, in 5.4.9.x? I think that once that time goes past six or seven hundred mlliseconds DS times out and detections won't happen.
IAmATeaf
Known around here
I’m not using DS, just ONVIF in the camera to trigger a recording which should then also add a timeline lightening bolt. With .6 everything is there except for the lightening bolt on the timeline.
sebastiantombs
Known around here
I'm getting alerts in BI from ONVIF detections but DS doesn't seem to analyze them. I've got to figure that one out.
I am pleased to report that when using 5.4.9.6, confirmed/cancelled for me are now more or less perfect. DS correctly cancels all the flickering shadows and night time reflections from rain, fog, spiders, bugs etc. No triggers are missed entirely (and regardless of the iframe interval).
One major reason is that the first image being sent to DS is always T+0 and not an uncontrollable T-xxx. It seems that checking “Begin analysis with motion-leading image” has no function. It makes no difference to the first image sent being at T+0. I did have the idea that it may be better to send the first image earlier at the trigger zone crossing (start of “make time”) but haven’t found a way to do this.
The second major reason is that DS now chooses (asterisk) the best valid image. The reason why it now does this is a mystery to me. I have not updated DS.
Checking/unchecking the “Pre-trigger video buffer” or setting ranges 300mS to 3S does not appear to influence the above comments.
Apart from the T+xxx msec, the % and the date, does anyone know what the other numbers in the DS .dat analysis mean?
Time to move on and try out the dark.pt thing if I can find out how. A link to somewhere or other would be helpful please.
One major reason is that the first image being sent to DS is always T+0 and not an uncontrollable T-xxx. It seems that checking “Begin analysis with motion-leading image” has no function. It makes no difference to the first image sent being at T+0. I did have the idea that it may be better to send the first image earlier at the trigger zone crossing (start of “make time”) but haven’t found a way to do this.
The second major reason is that DS now chooses (asterisk) the best valid image. The reason why it now does this is a mystery to me. I have not updated DS.
Checking/unchecking the “Pre-trigger video buffer” or setting ranges 300mS to 3S does not appear to influence the above comments.
Apart from the T+xxx msec, the % and the date, does anyone know what the other numbers in the DS .dat analysis mean?
Time to move on and try out the dark.pt thing if I can find out how. A link to somewhere or other would be helpful please.
@Dave Lonsdale - here you go:
github.com
Release DeepStack Dark Scene Objects Detection - Custom Model · OlafenwaMoses/DeepStack_ExDark
A DeepStack Custom Model for object detection API to detect objects in the dark/night images. It detects the following objects Bicycle Boat Bottle Bus Chair Car Cat Cup Dog Motorbike People Table ...
github.com
Many thanks wittaj!@Dave Lonsdale - here you go:
Release DeepStack Dark Scene Objects Detection - Custom Model · OlafenwaMoses/DeepStack_ExDark
A DeepStack Custom Model for object detection API to detect objects in the dark/night images. It detects the following objects Bicycle Boat Bottle Bus Chair Car Cat Cup Dog Motorbike People Table ...
Last edited:
Oh dear! I put dark.pt in the same folder as deepstack.exe and restarted the PC. Then I got my first ever 100/170 error. It‘s an i5-10500 CPU with 16GB RAM and SSD C drive. CPU % is low so perhaps there’s more to it than a shortage of resources.
Unfortunately my IT knowledge is pathetic so I’ve deleted dark.pt and restarted the PC. Fingers crossed BI/DS will work again as before…
Unfortunately my IT knowledge is pathetic so I’ve deleted dark.pt and restarted the PC. Fingers crossed BI/DS will work again as before…
sebastiantombs
Known around here
Create a directory under the DeepStack directory for your models. I was extremely original and use a directory named "models". Then put dark.pt in there.
OK sebastiontombs, I had another go using a models folder. The next trigger was "cancelled":-
So, not the result I was hoping for but apart from the Error 100, I also noticed two other things:
1. The timing of the images now ignores my setting of 5 x 250mS
2. An image at T-xxx has returned
3. Something I never realised which is the image displayed in the analysis corresponds with the one selected in the list details
At least I've now learned where to put dark.pt when the wrinkles are ironed out - thanks.
Just seen 5.4.9.7 is out. Can't resist....
So, not the result I was hoping for but apart from the Error 100, I also noticed two other things:
1. The timing of the images now ignores my setting of 5 x 250mS
2. An image at T-xxx has returned
3. Something I never realised which is the image displayed in the analysis corresponds with the one selected in the list details
At least I've now learned where to put dark.pt when the wrinkles are ironed out - thanks.
Just seen 5.4.9.7 is out. Can't resist....
sebastiantombs
Known around here
Sorry, the download has failed three times. My broadband is poorly this evening. I’ll try rebooting the router.
Rebooting the router fixed the Internet connection but for me, 5.4.9.7 didn’t stop the BI/DS errors. But, it’s dark and fog has descended. All my cameras are complaining and task manager shows the CPU spiking to 85%.
sebastiantombs
Known around here
I went ahead and "too the jab". Everything here is still running fine from what I can see. Virtually 100% detection, fast detection times under 200ms and no CPU spikes but I'm running the GPU version. I didn't try dark.pt, yet but may do that later this evening.
sebastiantombs
Known around here
CrazyAsYou
Getting comfortable
sebastiantombs
Known around here
avatar42
n3wb
Last weekend I noticed I was not on the latest version (5.4.9.6). After updating I seem to be getting almost no objects IDed on one box but the other seems to still work fine. Cancelled alerts say DeepStack -1. Tried rebooting and such.
Updated to 5.4.9.7 soon after it became available.
Still almost nothing getting IDed. Running a clip of me returning from the mailbox registered my face and nothing else. I've unchecked custom models, "default objects and face recog but that did not help. Cancelled alerts say DeepStack -1. Tried rebooting and such.
Last night, I finally downgraded the troubled server to 5.4.8.2 which seemed to get things working in analyse again but I still only saw 4 IDs overnight. Looking at videos in analyse there should have been the more normal of at least 10 times more. (I have a cam pointed at a heavily visited feeder with others that should see critters coming and going). The 1000+ cancelled alerts since the downgrade are back to saying "nothing found". Still digging but I'm starting to think I'm going to have to reinstall BI to sort.
My other BI5 server seems to be working fine on 5.4.9.6. Of course the one not working right is the more important of the 2.
Wish there was better logging / debug options to work with so I provide useful feedback other than it is not working. As a developer I know I hate just getting "it is broke" as a description. To fix a bug you generally need enough info to dup the problem.
Also going to need way to filter by objects seen at some point. It is an obvious feature oversight.
Updated to 5.4.9.7 soon after it became available.
Still almost nothing getting IDed. Running a clip of me returning from the mailbox registered my face and nothing else. I've unchecked custom models, "default objects and face recog but that did not help. Cancelled alerts say DeepStack -1. Tried rebooting and such.
Last night, I finally downgraded the troubled server to 5.4.8.2 which seemed to get things working in analyse again but I still only saw 4 IDs overnight. Looking at videos in analyse there should have been the more normal of at least 10 times more. (I have a cam pointed at a heavily visited feeder with others that should see critters coming and going). The 1000+ cancelled alerts since the downgrade are back to saying "nothing found". Still digging but I'm starting to think I'm going to have to reinstall BI to sort.
My other BI5 server seems to be working fine on 5.4.9.6. Of course the one not working right is the more important of the 2.
Wish there was better logging / debug options to work with so I provide useful feedback other than it is not working. As a developer I know I hate just getting "it is broke" as a description. To fix a bug you generally need enough info to dup the problem.
Also going to need way to filter by objects seen at some point. It is an obvious feature oversight.
kklee
Pulling my weight
5.4.9.7 is working fine for me. DS is working for detection.
I think Deepstack -1 means that Deepstack isn't running (or BI can't talk to it). You should be able to confirm this by clicking on the Test link on the AI configuration page.