Thanks for the proposal David.

I have two comments that I'd like to make:

First, given Gilles charts, does it still make sense to have 3 categories?
Or if we do have more than one category, could we change them so they are
not so closely related, and only grade the contest on one of them? (i.e.
the other one(s) are useful for ranking in terms of that category but don't
influence the outcome).

Several categories come to mind along those lines:

  o 'eye candy' factor
  o hof-worthy [this would only be useful for POV-Ray images of course]
  o educational value [from description or source code]

By 'educational value' I don't mean socially educating but value in
educating users of the tool in question, or of 3d graphics in particular.
For example a person who regularly enters, doesn't win, but over time
accumulates a particularly high average 'educational value' score could be
pointed out for special honor at some point, or something ... basically
these other categories are a method of tracking 'interesting' information
that may be one day useful to the admins, as well as allowing those viewing
the rounds to sort by that category if they so wish.

Secondly, for each round in which more than (say) 10 people vote (including
panel judges, and hopefully there will be at least several of those), I'd
like to suggest that the N highest and lowest votes for each image are not
counted. For # of voters 10-15, N might be 1. For 16-30, N could be 2, and
so forth. So if we have 30 voters, the two highest and two lowest votes for
each image are disregarded - so the final score is calculated from the
remaining 26 votes.

The intent of this is to attempt to offset the influence of votes that are
well outside the norm for an image. I recognize that it's rather crude, and
a better approach to looking for voting anamolies would be to e.g. view a
standard deviation graph or some such, but that's something a statistician
would be better qualified to talk about, rather than me.

IIRC (not 100% sure but I'm fairly positive - I'd have to check the code to
be 100%), when calculating the POVCOMP votes, we did exclude the highest
and lowest for each image (even though all votes came from a panel).

-- Chris

Post a reply to this message