Just wanted to say thank you to everyone for assisting me with setting up Deepstack on my computer. I appreciate all your responses and for taking the time to explain certain aspects of setting up the software and how to best configure it. I was able to get everything set up this past week and it is running perfectly. Still need to make some more adjustments on the camera's motion detection within Blue Iris, but otherwise working as it should. Truly appreciate all the responses and taking the time to assist me.
IP Cam Talk Custom Community DeepStack Model
- Thread starter MikeLud1
- Start date
What mode are you using, if you are not using High try it it might improve the detection also you might want to increase the contrast setting in the camera.
A good way to tell if you have enough contrast is to have a black object in the FOV and if it looks gray you need more contrast. If the object is gray increase the contrast small increments until it looks black but not to much that the rest of the image is to dark.
Last edited:
Thanks for the suggestions. I was on AI-mode "medium" and switched it to "high". Now, I am pretty sure I have the BI/Deepstack setup properly because I see in the BI-AI-log window under animal mode: bird, cat (see images), but never a raccoon. I guess that leaves improving my image quality by tweaking the contrast or a new cam.
I would also recommend unchecking Default object detection. The animal and combined model should detect all what you need.
I've been quite happy until now with the default DeepStack settings in Blue Iris, but I need to do something with our deer problem. I believe I've configured BI correctly to use MikeLud's combined.pt (placed the file in the MyModels folder). From what I can see the combined model is being used, but the deer is detected as dog/cat/bird. Is there something else I should check regarding settings in BI, or is it the camera that needs to be tweaked? Detecting persons etc. is working flawlessly.
Attachments
@Freddan101 Are you using "High" mode as @MikeLud1 showed above. AI detection of animals is not perfect. For example, two nights ago AI thought a person was outside. It was a spider instead.
kc8tmv
Getting the hang of it
I have the same problem with deer! Combined, dark and animal models ALL have no idea what a deer looks like for me. Using “high” mode also.
Deer are identified as dog, cat, fox, anything except deer.
My thought is simply there are not enough deer images in the models.
Sent from my iPhone using Tapatalk
One of the reasons is image count when I trained the model I only had about 50 images. I do have more images of deer's to add to the models.
The next updates to the models I am going to only have 3 labels in the combined model "person, vehicle, & animal". For the animal label I will use all types of animal to train the model. I will keep the animal model with individual labels.
kc8tmv
Getting the hang of it
Please don’t take my comment as a “slam”! You just had what you had! Thanks for the hard work.
Sent from my iPhone using Tapatalk
The person that gave me some more deer images made a deer only model, this model was trained with about 600 images. If anyone wants to try it out see attached.
kc8tmv
Getting the hang of it
I am including that one now. Is the label "to confirm" simply lowercase "deer" in that model?
The link with more details on the deer model. The label "to confirm" is lowercase "deer"
forum.deepstack.cc
Detect more common animals that are not in a zoo
I’m using .9 currently but I’m not sure it’s much better. I haven’t have any plant damage all summer, but I’ve noticed it takes a good while to get a confirmed hit. The deer are sometimes in the garden. When I run tests using the the dev kit and python, all my images are nailed quickly. My...
forum.deepstack.cc
Yes, saw the post and tested to set to "High" but same result I'm afraid.
Excellent, thank you. I will try it out immediately. If you need any more deer images, just let me know.
kc8tmv
Getting the hang of it
Post a screenshot of the AI analysis details if you have it turned on.
kc8tmv
Getting the hang of it
kc8tmv
Getting the hang of it
Try creating a clone and just running one model "deer". It might be an issue running four models, otherwise everything looks good.
Aha! I've been trying to get the combined model working on a Jetson Nano for a while now. I finally see the Blue Iris talking to Deepstack.
Deepstack shows 500, running the combined model, Blue Iris alerts, but just shows "nothing found". No errors! I tried person, Person, but I didn't try "People". I'll try that tomorrow.