Too many categories in image_manual - on super-labels and labels

I'm using the TACO dataset that has a few super-labels and a lot of fine grained labels.

If I use image_manual with all the fine grained labels the UI is just overwhelmed with the label count and the image won't show (see below).

I'm not sure how to best create more annotations of the same format using Prodigy.

For now, I am thinking I would use two distinct workflows :

  1. first create the polygon with a superlabel
  2. then break an annotated super-label with a multi-choice UI to add the fine-grain label

Or it could be nice to add a series of tags (fine grained labels) after a polygon is created

Hi! If that's the type of label scheme you're working with, displaying them all definitely seems inefficient. Also, drawing the boxes is already a complex enough task, so if the annotator also has to think about all those very fine-grained distinctions for different objects (plastic container or utensils? polypropylene vs. single-use bag?), this can really slow down the process. It's usually pretty obvious that there's some kind of plastic-y object and where it is, so you might as well lock that decision in quickly and take care of the specifics later. The specifics are also where you may have more conflicts and ambiguity.

So one possible workflow could be:

  1. Use the top-level super label only when drawing the boxes and get the boxes in quickly.
  2. Make another pass over the annotated data and show one object at a time, with a choice block with multiple-choice options for the sub-labels. Dependin on the category, you could allow one or multiple selections here. You could also use the image interface as the first block instead of image_manual, so the boxes can't be edited and you're just focusing on the sub-categories.