Corrected classes lose underscore

[kamilzyla]

Message from Robbie:

Something minor I noticed but it does cause issues is that the AI classifier labels the species names like this "Hog_Red_River" with underscores between words, but when correcting labels the underscores are lost. This means if you correct a label it creates a second class for the same species, so you end up with "Duiker_Blue" and "Duiker Blue" in the same database. This causes inconsistencies when I'm doing analyses that are easy for me to fix but will be a problem for others.

Steps to reproduce

1. Load a CSV file into the "Explore results" tab.
2. Override prediction for any photo with a_b string and export the results.
3. Load exported CSV and export it again.

Expected behavior

After the second export, the CSV file should have a_b string in both the common_name and label columns.

Actual behavior

After the second export, the CSV file has a_b string in the common_name column, but a b (without the underscore) in the label column.