Cumulative image count for best class probability in 2D classification

Is the cumulative number of images over best phiC correct? Shouldn’t e.g. the 0.8 bar and the 80% threshold have the same number of particles?

Hmm… yes this does look like something is wrong… do you see this all the time or only in this case?

It appears to always happen in a particular dataset. Other datasets I’ve looked at seem fine.