I agree that a competent write-up on the (true) theory of fashion seems to be missing. The usual way to deal with such situations is to act like the neural network you are: find some big dataset of [clothing example, fashionability analysis] pairs, consume it, then reverse-engineer the intuitions you’ve learned. If there’s no extant literature on the top-down theory available, go bottom-up and derive it yourself. (It will be time-consuming.)
Left a long comment that largely agrees, but I want to add: Do not just passively consume data! Working with a huge unlabeled dataset can be useful when you know nothing about the structure of the data, but you do know something about the structure of your data. You don’t have to understand fashion that well to build a good curriculum, or to give yourself a better architecture, or to add some side terms to your loss function. As with real neural networks, if you know anything at all about your domain, then you will get faster results by incorporating that knowledge into your training structure.
Honestly, not to sound like a fop, but if you trust yourself to moderate opinions you hear, go for something “old fashioned” like https://www.gentlemansgazette.com/ and learn some of the old-timey rules and then work forwards from there. Like many squishy disciplines, modern fashion is hopelessly complicated with 15 levels of counter-counter-counter-counter-counter-signaling, and going back to the basics can help.
Try to learn things like:
How do I know if a shirt fits?
What’s the basic mold than men’s outfits were built around (hint, it’s the suit) and how does the shadow of this hang over modern fashion?
Why do we care about “clashing” outfits, and how do I know if two pieces of clothing “clash”? What colors go together, what patterns go together, etc?
What materials exist, and how do I, personally, feel about including them in an outfit?
I agree that a competent write-up on the (true) theory of fashion seems to be missing. The usual way to deal with such situations is to act like the neural network you are: find some big dataset of [clothing example, fashionability analysis] pairs, consume it, then reverse-engineer the intuitions you’ve learned. If there’s no extant literature on the top-down theory available, go bottom-up and derive it yourself. (It will be time-consuming.)
Left a long comment that largely agrees, but I want to add: Do not just passively consume data! Working with a huge unlabeled dataset can be useful when you know nothing about the structure of the data, but you do know something about the structure of your data. You don’t have to understand fashion that well to build a good curriculum, or to give yourself a better architecture, or to add some side terms to your loss function. As with real neural networks, if you know anything at all about your domain, then you will get faster results by incorporating that knowledge into your training structure.
Do you know of such a dataset? I’ve tried spending some time on Pinterest for “fashionable” outfits but didn’t find that productive
Honestly, not to sound like a fop, but if you trust yourself to moderate opinions you hear, go for something “old fashioned” like https://www.gentlemansgazette.com/ and learn some of the old-timey rules and then work forwards from there. Like many squishy disciplines, modern fashion is hopelessly complicated with 15 levels of counter-counter-counter-counter-counter-signaling, and going back to the basics can help.
Try to learn things like:
How do I know if a shirt fits?
What’s the basic mold than men’s outfits were built around (hint, it’s the suit) and how does the shadow of this hang over modern fashion?
Why do we care about “clashing” outfits, and how do I know if two pieces of clothing “clash”? What colors go together, what patterns go together, etc?
What materials exist, and how do I, personally, feel about including them in an outfit?