Active learning to tune the object recognition / face detection online

It would be great to have a way in Blue Iris to mark false positives or false negatives to improve Deepstack’s model. For example, I have some cameras that would sometime detect my dog or some objects as a person. Instead of going through the work of actually training the model with several images from the different cameras in an offline fashion, perhaps it would be possible to have an API to tune the model automatically.
I don’t know if this is possible, nor whether “Active Learning” is the right terminology, that’s why I explained the whole Blue Iris scenario. If such an API would exist in Deepstack, perhaps the community or the Blue Iris developer himself would add this convenient functionality.

Hi @juliandom , online learning is not available in DeepStack at the moment, and we don’t plan on implementing such anytime soon. To ensure you are using a model best tailored for the scenes and images you are generating from your camera(s), you might want to try the approach mentioned by @shallowstack in the post below.

1 Like