Very cool blog post by @strickvl on using Prodigy to create training data for an object detection model for redacted text
I also really liked the section on annotation strategies and dealing with interesting edge cases – and why it's so important to get a feeling for your data.
As an aside, this is why annotating your own data is so valuable. You get to see the limits of the annotations, and you get to really own the decisions that are being made. It is a bit early for me to know which approach is the best solution to these two problems, but being aware of them is important.