IP Cam Talk Custom Community DeepStack Model

[MikeLud1]

Overnight I tried to test running Combined v2.0 and Dark. Running those two models together still didn't play nice. Today I'll try running Combined v2.0 and the standard Objects model together and see if it's a Dark model issue.

@sebastiantombs You said Combined v2.0 and Dark run fine for you. Would you mind sharing screen shots of your main AI settings and the Trigger > AI settings for your cameras? Thanks!

@105437 if you are using combined you do not need objects unless you what to trigger on the below objects. The main purpose for the new models is to remove all this junk

airplane, train, boat, traffic light, fire hydrant, stop sign, parking meter, bench, elephant, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork, knife, spoon, bowl, banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake, chair, couch, potted plant, bed, dining table, toilet, tv, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear, hair dryer, toothbrush

 

[105437]

@105437 if you are using combined you do not need objects unless you what to trigger on the below objects. The main purpose for the new models is to remove all this junk

airplane, train, boat, traffic light, fire hydrant, stop sign, parking meter, bench, elephant, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork, knife, spoon, bowl, banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake, chair, couch, potted plant, bed, dining table, toilet, tv, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear, hair dryer, toothbrush

Thanks. So the training and accuracy info is the same between the Objects and Combined models for things like car, truck etc?

 

[MikeLud1]

Thanks. So the training and accuracy info is the same between the Objects and Combined models for things like car, truck etc?

Combined is better.

 

[105437]

Combined is better.

Excellent, so objects:0 for each of my Profiles will be the way from now on. I still would like to work out the ability to run Combined and Dark together for a period of time. Ultimately, it would be great if Combined could evolve to include more training with dusk and dark images.

 

[MikeLud1]

Excellent, so objects:0 for each of my Profiles will be the way from now on. I still would like to work out the ability to run Combined and Dark together for a period of time. Ultimately, it would be great if Combined could evolve to include more training with dusk and dark images.

What mode are you running, High would be the best for all the new custom mode. I do not think there will be any speed impact if so a small amount.

 

[105437]

What mode are you running, High would be the best for all the new custom mode. I do not think there will be any speed impact if so a small amount.
View attachment 112281

Thanks @MikeLud1 Here's what I'm currently using. I'll try High and see how things work.

 

[MikeLud1]

I just posted animal V2.0 in the first post. I am not sure how well the new animals will work due to the low image count. Please let me know how V2.0 works for you.

I figured out how to download the images that DeepStack used to train the object model. With these images I can start to make the community custom DeepStack model, below are the steps to create the custom model.

The first step would be to create a new DeepStack custom model using the same image that DeepStack used but removing all the labels we do not want. Below is a list of the labels that I think we should keep and a list of labels we should remove. Please let me know if I should change the lists.

The second step would be to start adding new labels with images that everyone contributes to the custom models.

Thanks to @105437 for contributing images.

Combined Labels	General Labels	Animal Labels	Labels removed from Deepstack original model
person, bicycle, car, motorcycle, bus, truck, bird, cat, dog, horse, sheep, cow, bear, deer, rabbit, raccoon, fox, skunk, squirrel, pig Future Labels: coyote, possum	person, bicycle, car, motorcycle, bus, truck	bird, cat, dog, horse, sheep, cow, bear, deer, rabbit, raccoon, fox, skunk, squirrel, pig Future Labels: coyote, possum	airplane, train, boat, traffic light, fire hydrant, stop sign, parking meter, bench, elephant, zebra, giraffe, backpack, umbrella, handbag, tie, suitcase, frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket, bottle, wine glass, cup, fork, knife, spoon, bowl, banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake, chair, couch, potted plant, bed, dining table, toilet, tv, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, sink, refrigerator, book, clock, vase, scissors, teddy bear, hair dryer, toothbrush

​

Combined V2.0 Training Results:​

View attachment 112160View attachment 112161​

​

General V1.1 Training Results:​

View attachment 111835View attachment 111836​

​

Animal V2.0 Training Results:​

View attachment 112282 View attachment 112285​

​

 

[sebastiantombs]

@MikeLud1 Here you go. I just included one camera worth of day and night settings. I'm running, probably, 20 cameras with DS in total. All are set up on the same template though.

 

[MikeLud1]

@MikeLud1 Here you go. I just included one camera worth of day and night settings. I'm running, probably, 20 cameras with DS in total. All are set up on the same template though.

View attachment 112318

View attachment 112319

View attachment 112320

@sebastiantombs I just switched back to general after testing combined overnight. I am running general only during the day and night on all 11 of my my cameras. I do not have much wildlife here in Brooklyn just some racoons, opossums and lots of feral cats. On all my cameras I am triggering with IVS rules sometimes the IVS rule triggers on the feral cats and DeepStack cancel them out and sometimes DeepStack detect them as a person.

 

[sebastiantombs]

I'm using BI motion detection primarily. The camera I posted is a clone so IVS won't work very well with it, if at all.

I'm using combined because we do get a fair amount of wildlife coming by. That camera is for traffic but it will catch deer as a dog or horse as well.

I'm running kind of late today, slept late for a change, so I haven't had time to spend getting the animal model installed. I'll do that in a little while and implement it on cameras that don't send notifications for testing purposes.

 

[MikeLud1]

I'm using BI motion detection primarily. The camera I posted is a clone so IVS won't work very well with it, if at all.

I'm using combined because we do get a fair amount of wildlife coming by. That camera is for traffic but it will catch deer as a dog or horse as well.

I'm running kind of late today, slept late for a change, so I haven't had time to spend getting the animal model installed. I'll do that in a little while and implement it on cameras that don't send notifications for testing purposes.

When gather the image for the custom models make sure there are no DeepStack boxes.

 

[MikeLud1]

I'm using BI motion detection primarily. The camera I posted is a clone so IVS won't work very well with it, if at all.

I'm using combined because we do get a fair amount of wildlife coming by. That camera is for traffic but it will catch deer as a dog or horse as well.

I'm running kind of late today, slept late for a change, so I haven't had time to spend getting the animal model installed. I'll do that in a little while and implement it on cameras that don't send notifications for testing purposes.

You can also use the new animal model on a clone to capture the images for the labels I added.

 

[sebastiantombs]

Yes, I'll add animal to some of the clones that don't send notifications. While I like to see the wildlife activity I don't want alarms triggering and my phone triggering all darn night. I really value my sleep

 

[MikeLud1]

The camera I posted is a clone so IVS won't work very well with it, if at all.

If you want to use IVS what you can do is have the master trigger the clone.

 

[sebastiantombs]

Or just use DS on the master, which is what I do. That lets me easily discriminate between them.

Alternately, the clone can also be triggered by ONVIF to achieve the same thing.

 

[eeeeesh]

Coyotes? Ya, we have a coyote problem - this was my backyard after they jumped the front 6' wall

 

[105437]

Coyotes? Ya, we have a coyote problem - this was my backyard after they jumped the front 6' wall

View attachment 112332

Would love to see coyotes added to the combined model. We have them here every so often. Not sure how DS will tell the difference between a coyote and dog. If it can, that would be impressive!

 

[105437]

@sebastiantombs Thanks for posting your settings. Mine are pretty similar except I had "Use main stream" selected. I have deselected that to see if I can successfully run the Combined and Dark models together.

 

[sebastiantombs]

@105437 I've found using the main stream to be counterproductive when using DS. YMMV

 

[MikeLud1]

High would be the best for all the new custom mode.

Just to let everyone know what the High, Medium, and Low settings do, below is a part of the code from DeepStack. If you set mode to High DS will downsize the image that BI sends it to 640 x 640 before it tries to detect what object is in it. Medium downsize = 416 x 416 and Low downsize = 256 x 256. All of the training I am doing is at 640 x 640. The original DS object model is also trained at 640 x 640. Below are also samples of image from the general training.

DETECTION_HIGH=640,
DETECTION_MEDIUM=416,
DETECTION_LOW=256,
FACE_HIGH=416,
FACE_MEDIUM=320,
FACE_LOW=256,