We are seeing cases where reviewers find that an example image is ambiguous enough to justify exclusion from our dataset. As an example, consider a cats and dogs classification project with an example that the reviewer cannot unambiguously state is cat or dog. We want to remove it from the dataset altogether.
We do not, however, want to grant data row deletion privs to our reviewers as this would be a significant escalation in privilege. Instead we want to flag the data row as “to be deleted” and clean it up later (delete it from the lb project as well as from our managed storage resource simultaneously) via service account. What is the happy path for achieving this?
I’ve looked in to workflows but the states of “done” or “rework” don’t seem to allow this. If we could use logic to add a tag that would be sufficient, i.e., the reviewer sends the label to a special “to be deleted” logical step that appends a metadata tag to the datarow flagging it for deletion before moving it to “done”, we could then scan for datarows with the specific tag.
