14
\$\begingroup\$

This is a question about what counts as too far towards an art contest.

If I asked the question "render the Mona Lisa in 1000 bytes" on the main site, would it be on topic? Assume the question contains precisely specified guidelines about what the program is and isn't allowed to do, but that the winning criterion is "popularity contest", with the expectation that people will vote for the answer whose output most closely resembles the Mona Lisa. The only restriction on entries is that the rendering part of the code cannot be more than 1000 bytes in length.

Note that I'm not really asking whether this would be a good challenge, though I might post it if it's on topic. This is more about understanding what the community considers acceptable winning criteria for questions. I'm also not asking whether it should be on topic, but rather whether it de facto is on topic according to current rules and conventions.

I ask because many existing on topic, well recieved questions have "looking like a given input image" as the goal, with American Gothic in the palette of Mona Lisa: Rearrange the pixels perhaps being the cream of the crop. I really like these questions, and I want to know how far that kind of idea can be played around with without straying into "art contest" territory. Answers backed up with reasoning and/or links to previous meta discussions would be appreciated.

Edit: I've put a draft of the proposed question in the sandbox.

Edit 2: The question has been posted on the main site, where it was closed, reopened and then closed again. I have now edited it based on feedback given in various places. (Here, in the comments on the question, in another meta post about popularity contests that was sparked by my question, and in a chat session with moderator Dennis.)

\$\endgroup\$
  • \$\begingroup\$ The crux of the question (in my point of view) is: Are we allowed to use popularity-contest to judge the accuracy of submissions? \$\endgroup\$ – Nathan Merrill Jan 10 '16 at 16:00
  • \$\begingroup\$ I find the answers below and the votes on them to be pretty confusing - I've decided to just post it and see what happens. \$\endgroup\$ – Nathaniel Jan 12 '16 at 12:03
  • \$\begingroup\$ Does anyone else think it would be useful to have a new separate meta question for discussing the challenge now that it has been posted? \$\endgroup\$ – trichoplax Jan 23 '16 at 17:25
19
\$\begingroup\$

Yes

But I think might not be ideal.

How I would do it is choose some specific image of the Mona Lisa that everyone has to aim for, then give each submission a score based on how different it is from this image.

This score could be the number of pixels that are the exact correct color (higher score wins). Or if you want more granularity, have the score be the total of the absolute differences between the RGB channels across all pixels in the goal and submission images (lower score wins).

This way it would be a instead of , and you wouldn't run into any "art contest" issues.

I personally would not be opposed to a just as you suggest, since a more "human" scoring method, rather than just comparing pixels, might make the results more interesting. But some people on this site probably would call it an art contest, which is why I'm saying the is safer.

\$\endgroup\$
  • \$\begingroup\$ I think the objective winning criterion would actually make it much less interesting though. For example, in answering your excellent "rearrange the pixels" challenge, people came up with some really novel dithering-based solutions. If used absolute difference from a target image as the winning criterion then there would be no need to do that (it would be sub-optimal), so the winning solutions would look posterised and uninteresting. \$\endgroup\$ – Nathaniel Jan 10 '16 at 5:36
  • 1
    \$\begingroup\$ @Nathaniel I agree. Again, I personally am not against the pop-con idea, in fact I think it's a pretty good idea - but some folks around here are fussy about "art contests" which is why I call the code-challenge safer (i.e. less likely to be closed). \$\endgroup\$ – Calvin's Hobbies Jan 10 '16 at 5:40
  • 1
    \$\begingroup\$ In the field of image processing, is there some standard metric for visual similarity? I could imagine there's been a lot of research effort put into this. \$\endgroup\$ – xnor Jan 10 '16 at 6:36
  • \$\begingroup\$ I think it is a good thing that people are fussy about popularity contests, but I don't think that's a reason not to try. It's easy to get a pop con wrong, but that just means you should put more time into getting the question ready before posting, and for most people, get plenty of feedback too. \$\endgroup\$ – trichoplax Jan 10 '16 at 17:58
  • 6
    \$\begingroup\$ Absolute difference is RGB is a horrible metric for any purpose. A true perceptual difference based on XYZ would be overkill, but it wouldn't be too hard to specify an approximate perceptual difference metric based on HSV or HSL (measuring distances within the cones rather than direct L_alpha on the three-tuples). \$\endgroup\$ – Peter Taylor Jan 10 '16 at 19:20
  • \$\begingroup\$ @PeterTaylor I think any per-pixel measure would not be ideal, since colour perception is very dependent on the surrounding context, and perception of texture, depth etc. are fundamentally non-local. (But actually, the more I think about it, I think that trying to get "objectively" close to the right pixel values would make an interesting challenge, even if the results don't actually look that much like the original picture.) \$\endgroup\$ – Nathaniel Jan 11 '16 at 3:10
  • \$\begingroup\$ @Nathaniel At a minimum, use the sum of the r² differences. \$\endgroup\$ – Riking Jan 19 '16 at 20:14
2
\$\begingroup\$

Answer for voting:

Yes, it's on topic, in its current form as a . It might not be ideal, but it's within the rules as they're currently set out and practised.

(I'm adding this answer because it's been suggested to me that the votes for Calvin's Hobbies' answer are because people agree that a is not ideal, rather than because they agree that the idea as I proposed it is on topic. Therefore, please vote for or against this answer according to whether the idea as I proposed it is on topic without changing it to a more 'objective' winning condition. Please note that this is purely for voting about whether it's on topic, and not whether it would be improved by changing it.)

\$\endgroup\$
-6
\$\begingroup\$

No.

A spec that simply said "recreate the Mona Lisa in 1000 bytes" would be far too broad and vague, even for a . Without an objective criteria for what makes an image resemble the Mona Lisa, golfers have no real goal to shoot for (aside from just simply displaying a photograph of it) and voters have no real guiding criteria for what answers to vote for.

Just because it is a doesn't mean there shouldn't be any objective criteria for what makes a good answer.

EDIT: Per Calvin's suggestion, I think the best idea for this is to turn it into a (perhaps some sort of Levenshtein distance-esque scoring).

\$\endgroup\$
  • \$\begingroup\$ Ok, but I guess my question is why is it considered broad? There are many other graphical-output popularity contests that have criteria that seem to me much vaguer than looking like the Mona Lisa. Some are closed and some are not, and I would like to understand the reasoning. \$\endgroup\$ – Nathaniel Jan 10 '16 at 5:17
  • \$\begingroup\$ Ideally there shouldn't be any graphical output challenges like that. Links? \$\endgroup\$ – a spaghetto Jan 10 '16 at 5:18
  • \$\begingroup\$ Wait, so, if I challenged to recreate the Mona Lisa to a 16 dpi (from actual size) without compression and only using some form of algorithm and asked them to golf it down, would that be on or off topic? (this is to verify what the answer is saying) \$\endgroup\$ – Addison Crump Jan 10 '16 at 5:20
  • \$\begingroup\$ @quartata "Assume the question contains precisely specified guidelines about what the program is and isn't allowed to do..." Seems like the spec is not simply saying "recreate the Mona Lisa". \$\endgroup\$ – Calvin's Hobbies Jan 10 '16 at 5:32
  • \$\begingroup\$ @Calvin'sHobbies just to make sure we're all answering the same question, I meant that the spec would not contain any ambiguities about whether you're allowed to load data from an external file, etc. The goal would just be "recreate the Mona Lisa" and the main restriction would be the code size. \$\endgroup\$ – Nathaniel Jan 10 '16 at 5:37
  • \$\begingroup\$ @Nathaniel All of these appear to be fine (perfectly objective criteria) except for the black and white forest one (which does have some comments on it stating it's more of an art contest). Something you do have to keep in mind when looking at old challenges is that site standards have changed over the years; some challenges that were fine then would not be fine now. \$\endgroup\$ – a spaghetto Jan 10 '16 at 6:07
  • 1
    \$\begingroup\$ @quartata I'm not sure why you say those are fine but my proposal isn't. The mathematical art one is a pure pop con with no judgement criteria; rearrange the pixels is literally the same as my "look like the Mona Lisa" criterion, except that the input image is a parameter; images with all colours is a pure pop con; paint with numbers is a pure pop con. What is it about what I proposed that's different from any one of those? (I do know that standards have changed over time; that's part of the reason for asking this question.) \$\endgroup\$ – Nathaniel Jan 10 '16 at 6:28
  • \$\begingroup\$ @Nathaniel Because all of those have objective standards (the goal to make a picture with all colors, for instance), whereas what the Mona Lisa looks like is purely a subjective standard. \$\endgroup\$ – a spaghetto Jan 10 '16 at 15:02
  • \$\begingroup\$ @FlagAsSpam That would be fine. \$\endgroup\$ – a spaghetto Jan 10 '16 at 15:20

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .