Classify more dice data

[belgiumkansas]

Currently our YOLO model sees both dice 5 and 6 as dice 6 most of the time. We need to be able to accurate separate these for the dice task and ideally want to avoid classical CV processing to figure this out.

We don't have data on the new dice obstacle so this will allow us to continue developing until that time.

[skhadem]

Closing due to new buoys taking priority