Hey, you are asking the right kind of questions! There are quite a lot of things which make Animal Detect unique, while we constantly build on new and improved features. 

Currently, we are working with ViT models and different methods and algorithms in combination with classical detection models, to add even more true positive detections into classes, while also using these algorithms to suggest classes for images.

Our algorithms scales and works best on larger uploads of data, where we actually create a model from the data, while we run classification. 

While all of this is happening, we also help to sort the detections based on similarities, for a quicker way to re-classify and handle human-in-the-loop. 

I am trying to (if it works) upload a gif, where you can see a result from where we got detections of over 21.000 animals from about 6.000 images. What you will see is that the images are not sorted based on the label (even if you can filter them as you wish) but instead similarities. This makes the reviewing and adjustment of classes a breeze, compared to manually going through and spotting a black-backed jackals together with a mix of other types of animals. 

We are always open for feedback to improve this even further 😊