After Merging Datasets, Re-balancing (Train/Val/Test) excludes multiple classes in VAL split

I agree - will this be fixed? The train val split feature is somewhat obsolete right now due to it being so imbalanced. Also, is there a way to re-split images after labeling is finished?