Perhaps someone will take this up in the future if development time is available. The current implementation of "Score All" is somewhat hacked together. As it stands creating a per-object class table causes predictions to be run twice because that optional feature was added at a later stage than the original script. The original code is somewhat messy and so there's not a straightforward way to generate both the per-image and per-object tables at the same time. It should be relatively simple to redesign the process to do everything in one pass, but it's a more major refactor than we had time for in 3.0.
Perhaps someone will take this up in the future if development time is available. The current implementation of "Score All" is somewhat hacked together. As it stands creating a per-object class table causes predictions to be run twice because that optional feature was added at a later stage than the original script. The original code is somewhat messy and so there's not a straightforward way to generate both the per-image and per-object tables at the same time. It should be relatively simple to redesign the process to do everything in one pass, but it's a more major refactor than we had time for in 3.0.