sebastiantombs
Known around here
@fenderman It was alerting apparently based on the image of our cars even though it was being triggered by a raccoon or fox. The alert was tagged "car 76%".
Hell Yeah! Direct Deepstack Integration - 5.4.0 - March 31, 2021
- Thread starter fenderman
- Start date
I think the last update 5.4.3 from just in the last few hours fixed tagging items in images its seen before.
My alerts cleaned up on their own, no changes from me. Its now only highlighting the new object in the alert image.
Perfect and super fast development!
You can see it tagged the red truck just one time after I did the update and then never again.
Edit: Here is the blurb from the patch:
DeepStack objects recognized only when substantially overlapping areas where Blue Iris is
also tracking objects in order to reduce the detection of static objects.
My alerts cleaned up on their own, no changes from me. Its now only highlighting the new object in the alert image.
Perfect and super fast development!
You can see it tagged the red truck just one time after I did the update and then never again.
Edit: Here is the blurb from the patch:
DeepStack objects recognized only when substantially overlapping areas where Blue Iris is
also tracking objects in order to reduce the detection of static objects.
Last edited:
sebastiantombs
Known around here
Ssayer
BIT Beta Team
Hah, I have a bunch of them! Also a CD (that I use as a flag) on top of one of my disk golf goals. Of course, I have min confidence at 3%, so there's that. 
sebastiantombs
Known around here
Yeah, I'm down at 1 to 3% depending on the camera.
I have noticed that the missed detection fall into two categories. One is dark colored vehicles, even during full daylight hours. The other is vehicles going too fast. By the time BI triggers and sends an image I don't think there's enough left in the frame to be identified by DS.
I have noticed that the missed detection fall into two categories. One is dark colored vehicles, even during full daylight hours. The other is vehicles going too fast. By the time BI triggers and sends an image I don't think there's enough left in the frame to be identified by DS.
My cameras are IPC-T2431T-AS 3.6mm from EmpireTech. They are generic Dahau 4MP fixed turrets.
4MP Starlight Turret IP Camera | IP Cam Talk Store
I'm not sure if this is related to others having the problems at night, however, the software is not picking up movement in the driveway with deepstack turned on when it comes to cars. It will pick up people but not the car driving it. I think that it's the headlights. What can I do to fix this? It's a DS-2CD2347G1-L 4 MP ColorVu Fixed Turret camera.
sebastiantombs
Known around here
You can try using High Light Control in the camera, if the camera has that and I'd expect a decent Hikvision to have it. You can also try lowering the "Confidence" level in DeepStack. I'm down at 1 on a couple of cameras in the night profiles. Even then it's not very reliable but YMMV.
tech101
Known around here
Hi, Does anyone know how do we actually train DeepStack for objects any easier way ? or a guide ? To my understanding atleast so far deepstack do not learn or get better Automatically you have to train it. Hopefully in Future it can do it in some automated fashion. Reason So I can train this for some of the objects it is having issues labeling correctly.
sebastiantombs
Known around here
I seem to have lost the entire database of clips. In "All Clips" very few show and those are tagged "jpg Group" from cameras that are in groups. Tried a repair/regenerate as well as delete/regenerate and neither worked. I've reverted back to 5.4.1.1 and still no joy but at least it's adding new alerts again. I'll email BI support later. Oh yeah, in the stats at the bottom of the main console scree it's showing 7921 clips.
Go to camera settings/alerts click "on alerts" and then add a new alert. You can now put in "person" for example and it will only fire alert when a person is detected. Works GREAT !
Attachments
Ah thanks. I interpreted it differently -- as in, I can search for alerts with "person" or "sandwich"
Now that he fixed static objects being flagged 100% of the time, it seems the only use for zones would be to reduce cpu usage (ds won't be analyzing hundreds of images every hour)
Zones are still great for defining a alert as well. For example, I can set a zone on my front porch by mailbox and make it alert me only if a person is in that zone...as in mailman. You've got mail LOL
The camera without deepstack picks up the cars entering at night. The driveway is lit by 3 lights so no infrared used at night.
Turn the confidence down to 30 and see what happens. Some have it as low at 5 so you can decrease it quite a bit to help pick up on the cars at night.
I just review my footage from last night and I see the same problem you have. If the cars are moving slow, the DS image captures the car but if they are moving fast, the captured image is blank, the car has moved out of the frame so DS says nothing there. I have it set at 2 frame capture and wondering if moving to 1 would help with nighttime captures. My confidence was set at 40. Going to test out confidence at 30 and 1 frame to see if it captures the cars at night.