AI assistants are what we call our AI tooling that you can use to automate parts - or all - of your annotation work.
The concept behind them is simple. Instead of relying on pre-trained
models or large, generalized models, we train models using the data you
provide. This way, we can guarantee that we will be able to deliver
assistance no matter what your use case is. Think about it as having
models custom-tailored for what you are working on.
We will train model(s) and activate the assistant(s) after you've
labeled 10 images and set the image status to "To review" or to "Done".
Which assistants we train depends on what data you put in. If you put in
image tags, we train our Image Classification Assistant. If you
annotate using bounding boxes, then we train our Object Detection
We will then retrain the assistant(s) whenever we've received 20% more
data. So, given that you have labeled 10 images, the next model will be
trained after 12 images. Then 15. Then 19. And so on.
Use for speeding up image classification.
Use to get predictions on potential metadata to add to your annotations.
Predicts the class of any annotations made so that you don't have to switch between classes manually.
Shows you predictions of where our model thinks bounding boxes should
be placed on an image that you can then accept one-by-one or accept all
at once. Used to speed up object detection.
Shows you predictions from our AI model on where you can add masks/polygons for your Instance Segmentation annotation task. As for Object Detection, you can accept them one by one or accept all.
The same as Instance Segmentation Assistant, but for Semantic Segmentation.