As part of the effort to provide a collection of sample custom DeepStack models trained on popular and custom datasets, we are happy to announce the release of the DeepStack_ExDark custom model for detecting people and 11 other common objects in dark/night images and videos. It is a YOLOv5x...
I loaded up the other custom models for logo and plates, so will see if that works when UPS or FedEx go by. It isn't picking up plates but it said they were trained for Norway, so no shocker there.
What I do not know is if BI is picking up on these custom models.
I did notice that in the DeepStack directory is a sharedfiles folder that had face model and a few others that I assume are being used by DeepStack for BI, so I put the custom models in there to see if anything pops up.
OK, we need someone more experienced with DeepStack and how it works and its integration with BI lol.
I ran the tuning option in playback and stopped it with the UPS truck right in the center of the image as that seems to pull up everything the DeepStack can identify whether you asked for it or not LOL. Deepstack identified my neighbors potted plants, the stop sign, mailbox, parked cars, a ball in the yard, the UPS truck itself, but not the logo. So it must be coded in BI which models to use and even putting these in the sharedfiles folder didn't work.
I also ran the command file to run all of these, but they do not appear to be accepted by BI.
Let's see if anyone here is more experienced with how this works. I will send an email in a day or so if we do not see anyone have success with this and see if we can add custom models in DeepStack for BI. Maybe that is a future update, but in my novice experience, it looks like it is simply adding the models into the folder and maybe he needs to tell BI to run DeepStack with any models in that folder?
I looked at the dark training set. Doesn’t seem that it would be that useful. I train two models: one with color images and the other with IR images. When the camera switches to IR it sends a gray scale image, and I use that to determine which model to use. When I train my models I first separate the dataset into color and gray scale. You can’t simply apply a gray scale filter to a color image for training, as an object lit by IR looks quite a bit different.
The other thing which helps a lot is detecting stationary objects at startup and excluding them. Not sure how hard that would be with BI/DeepStack.
@brianegge - since it sounds like you have dabbled with this, I have been trying to get a custom model to work in BI and it isn't working LOL but not sure if user error or if it is something within BI that the developer would need to add to allow custom models.
Maybe it is something else on BI's end that needs to be ran to allow additional models to work, or maybe there is a coding change I would need to make to a file in DeepStack? Is there anything within the DeepStack files that specifies the models that are used?
I can confirm that the custom models are running by running this command:
Yet for like the logo custom model, I cannot get the logo to show up in BI. I tried the new tuning option and nothing shows up - UPS and FedEx truck are as clear as day and it picks up it is a truck, but doesn't pull the logo.
I will try the custom model soon too. Will read thru the posts here thank you @wittaj. This is a game changer with custom models now BI has got so much better in past couple of months it was already so good and with these updates it has added so much more..
Does the deepstack models and deepstack itself in BI has to be on same port or can it be different ports ?
Reason I am asking seems like the custom model has a Command line I have to input which is port 80.
however my deepstack runs on port 82 which is in Blueiris. I can make that port 80 as well But was wondering if for any reason they need to be on same port or not.
Oh ok - that is if you run it outside of BI - I was trying those before the custom option became available.
I changed the port # to what it was in BI and they all showed running but never showed in BI.
I don't think we need to do this manual start as BI should start it now with the custom setup. I know at least one person got the logo and dark model to run.
Yep there's no need to manually start custom models outside of BI. I stopped Deepstack, put custom models in the folder and then hit start. BI took care of it all
Thank you @wittaj & @hikky_b I think its running as I can see a alert with an uppercase Car and lowercase car. I think I did read it somewhere here that dark custom model is the one with uppercase. Anyhow still see 404 Will see if openlogo one shows up or not.
FYI I DO see the normal DS detection indication of "person" on the image that was sent in the text message. Tried on 2 other cameras and the &JSON macro is still {"success":false,"error":"invalid image file"} on all of them.
I wonder what is up with mine? DeepStack works when I rollback to the stable version and when I update to the latest it doesn't work. I have tried deleting and adding cameras to no avail.
forum.deepstack.cc
github.com
