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 was investigating also another issue whats up with temp .rdb files with Deepstack mine is now taking almost 160 gb of diskspace. Anyone else seeing anything like this ?
When I have DeepStack running and interact with it directly (using a Python script), it seems like you need to call the custom model directly to get the correct results. For example, sending image data to the default API endpoint "http://deepstack_host:5000/v1/vision/detection" yields quite different results than sending the same image to "http://deepstack_host:5000/v1/vision/custom/dark".
Since the AI configuration page in BI only asks for an IP and Port, I can only assume it's using the default API endpoints for detection and not the custom dark endpoint. So even though your custom model is loaded, it doesn't seem like BI is sending data to it.
The page also asks you where your custom models are located and that is how it calls and pulls the custom models. You need to place them in its own directory.
The page also asks you where your custom models are located and that is how it calls and pulls the custom models. You need to place them in its own directory.
The page also asks you where your custom models are located and that is how it calls and pulls the custom models. You need to place them in its own directory.
I am having trouble getting custom models working. I've set the model folder with licence plates and dark models but not sure if im meant to specify "licence" under inside the custom text field. Can anyone advise?
If you want all models used, including the stock one, leave the custom models field blank in the AI config for the camera.
To specify a model list the name of the model. Multiple models can be listed and separated by commas.
Models can be avoided by listing the model followed by a colon and a zero, IE objects:0 Multiples can be avoided by listing them separated by commas.
No spaces are used between any of the punctuation.
Be careful to get the names correct, including capitalization if the model has any capitals. The case of letters must match, exactly, for DS to recognize them.
Don't forget to specify objects you want to detect for each model in use unless you want to search for and detect all objects in the model.
If you want all models used, including the stock one, leave the custom models field blank in the AI config for the camera.
To specify a model list the name of the model. Multiple models can be listed and separated by commas.
Models can be avoided by listing the model followed by a colon and a zero, IE objects:0 Multiples can be avoided by listing them separated by commas.
No spaces are used between any of the punctuation.
Be careful to get the names correct, including capitalization if the model has any capitals. The case of letters must match, exactly, for DS to recognize them.
Don't forget to specify objects you want to detect for each model in use unless you want to search for and detect all objects in the model.
Great explanation. Also, I'm not 100% sure... but it may be that Blue Iris needs to be restarted before DS model configuration changes in the Trigger Tab > AI take effect.
I should have added that it's best to shut down DS through the BI console before adding a model or making any changes to the model directory at all. The re-start DS from the BI console and confirm it is running again.
Knew I forgot something and couldn't think of what it was.
I should have added that it's best to shut down DS through the BI console before adding a model or making any changes to the model directory at all. The re-start DS from the BI console and confirm it is running again.
Knew I forgot something and couldn't think of what it was.
Wish there was more control over models and when they're used.
Right now I'm successfully running stock DS model and the extra dark one but I get the double labels for people and person.
Be nice to have a schedule of when models are used.
Wish there was more control over models and when they're used.
Right now I'm successfully running stock DS model and the extra dark one but I get the double labels for people and person.
Be nice to have a schedule of when models are used.
I appreciate the reply.
Specifically where would I be specifying dark:0 etc? In which settings box?
Right now I've just got custom models enabled and pointed blue iris to the folder on my desktop that holds the dark model.
I think I got it. Thanks!
So I'd run different trigger profiles for day vs night etc.
And in the trigger settings for each respective profile enable or disable the call to whichever model I want.
Only other thing I guess is figuring out exact model name for stock model. I assume custom model is just the name of the file?
This then allows you to tweak motion detection settings based on day and night as well. Most of my cams are the same and the night profile follows the day profile, but in some instances they are different to account for things shadows or what not and allow you to better dial in the settings.
This then allows you to tweak motion detection settings based on day and night as well. Most of my cams are the same and the night profile follows the day profile, but in some instances they are different to account for things shadows or what not and allow you to better dial in the settings.
Had some luck with trying this. My night profile was working correctly only using the dark model when I put "dark,objects:0" into the custom models box.
I was having issues during the day though where it was still using both models even though I had "dark:0,objects" in the custom models box.
It seems it acts differently depending on which one you list first. dark,objects:0 & objects:0,dark should be equivalent but it seems the software doesn't act that way?
My maintenance plan had expired so I upgraded to be able to update BI since I hadn't since December. Maybe that will help resolve the issue.
ipcamtalk.com
