• 0 Posts
  • 9 Comments
Joined 2 years ago
Cake day: October 8th, 2023
  • ijon_the_human@lemmy.worldtoLemmy Shitpost@lemmy.world•Choose a number, 1-5!
    2·
    28 days ago

    EDIT:

    The University of Edinburgh has a publically available visual dataset of 59 forks:

    https://homepages.inf.ed.ac.uk/rbf/UTENSILS/DINNER_FORK/dinner_fork.xml

    I guess one way to approach this would be to use something like Skyrim’s character generator. Not sure how challenging it would be to put together something like that but it’d be a laugh.

    Your suggestion about a 3d model with adjustable parameters is something I could do in Grasshopper. There is the problem of defining the parameters though. Not impossible by any means but determining which features are the most critical and which could be excluded could end up being surprisingly difficult.

    As far as copyright is concerned I don’t think we’d encounter problems when using product photos from Ikea and the like. But your right about the combination part. Not only do the variations need to exist, we’d also need comparable photos of relatively high quality.

    If we were to actually do this though, I think it would be best to start simple, do a sort of goldilocks prototype and see how it goes from there. For instance splitting ten (maybe even less) of the most popular forks into three parts and poll people on those (either separately or as combinations). If the number of initial fork designs is not too high, I could even model each of those separately to avoid copyright issues. Then again this works better for determining the preferred shapes and not as well for preferred scale – e.g. how long the tines are compared to the handle etc. I guess there would need to be two stages, shape and proportions. The proportions would be quite simple though and the initial images could just be scaled and skewed to achieve the desired look.

    Another idea would be to go through the features one by one, having a poll at each stage. Every time the result is split, we could “fork” the design into multiple ones. This would require a whole lot of orchestration though and I have a gut feeling this wouldn’t work very well in practise. This is also based on the assumptions that there is only one perfect design for each respondent and that the preference for features is not interdependent.

    Then again we could utilise gen AI to turn simple proportional sketches / illustrations (modifiable or user drawn) to more realistic images which maybe users could then rate.___

  • ijon_the_human@lemmy.worldtoLemmy Shitpost@lemmy.world•Choose a number, 1-5!
    2·
    28 days ago

    I have a couple of ideas but I don’t know which would be the easiest.

    We could analyse and break down forks to their design features and then let people mix and match. This would take a lot of time though.

    Alternatively we could scrape a bunch of fork images and build a large dataset. Then show two at a time to users and let them pick their preferred one repeatedly until we know the preference hieararchy for that user. Then compare users and divide them into demographics.

    I guess it would also be possible to have a visual model and some edit options like make different parts longer / thicker. I feel this is unnecessarily complicated.

    Another idea would be to break the fork into two or three parts. Not perfect but faster than exploring all design features. Tines, body, end for instance and let people either mix and match or do some for of a/b testing.

    My expertise is in design so I can help with those aspects but coding I unfortunately cannot do. Would be a really fun project designwise too…