Hacker News new | past | comments | ask | show | jobs | submit login
Taste graphs will transform fashion (chicisimo.com)
20 points by aldamiz 5 months ago | hide | past | web | favorite | 17 comments



I have something on the order of 20m^3 worth of clothes. I'm seriously weirded out by how none of the "fashion tech" companies do anything regarding appropriate outfits for the weather conditions. 50% of my outfit choosing time is spent figuring out whether fleece works for 0C-but-windy, what the right number of layers is for stuffy public transit, breezy outside, and chilly office AC, and whether something is at risk of majorly embarrassing malfunction if I run. I don't need an AI to have taste for me (thanks but I have it we can't all be clueless-and-nerdy...), I need it to have fast, appropriate decision-making in case my original plan falls through because of surprise snowfall.

You're welcome to make a startup based on that idea ;) up here in Canada where we have real seasons I'll find you some great test users!

(Although I've been to the Bay Area and it's not without weather challenges)


Fashion isn't based on personal taste. It's based on emulating other people and fitting in crowds.

It would be better to train a classifier to learn what clothes fit the current trend.


Pinterest has been doing this for years:

https://medium.com/@Pinterest_Engineering/taste-graph-part-1...

How is this different/better than what they are doing?


Hey Andrew, we started building it in 2012. The concept is similar as you say. The most relevant difference is the ontology: our ontology is very specific to what-to-wear needs, theirs isn't. Thanks


I wouldn't be so sure about them not having the "what-to-wear" ontology. I worked with them a few years ago and they had a pretty long and comprehensive list of these categorical webs. It's a core area of focus for them.

Just a heads up.


Thanks Andrew. They are an amazing team with a great ontology, I agree. Thanks for the heads up!


Wow. Got to be the worst fluff article I read in a while. Its just a repeat of the words taste graph every other sentence.


I agree it's pretty badly written, but I believe the idea is sound since I worked on something similar in my past startup.


I need a taste graph of your comment to understand if your taste graphs to our interested taste graph audience


You’ve missed Oncology.


That's a funny company. They patented an implementation of a Bayesian Network and call it something new and amazing.

Not that it matters, as I'm certain other companies use more modern techniques (deep learning) for "outfit advice".


Please don't be a jerk in response to other people's work. If you see a weakness or something missing, point it out in a way people can learn from. Just spraying bile on things makes this place meaner.


I appreciate your desire for a better hackernews community, but I disagree with your assessment of my comment. The article is clearly for marketing, not quality content. The original poster is indeed a member of the company. That's not to say company articles are all bad, but the merit of this article is what I based my comment on. But to be fair, what I consider "bile spewing" is based on reddit quality comments, so perhaps I should rethink my commenting style.


You're getting the right idea by the end of that comment :)

Bear in mind that HN began as a community of startup founders and enthusiasts, and even though the audience has broadened, it's still part of the culture to be charitable and supportive to people who are just starting out and trying to get something off the ground.

Not everyone entering the startup world has an MIT PhD (we hope!) and the ability to write content that has both technical depth and literary polish.

We all have to start somewhere, and the only way to start getting better at something is to put something out there and see what comes back.

Critique is fine, but we can be constructive and encouraging to help people improve - just as we'd hope people would to for us if we're taking a risk on trying something new.


Its been nearly two months since I started emailing hn@ycombinator.com without a reply. I’ve changed email providers in case it was a spam issue, and still nothing. So, I’m trying this again. Please respond.


Thanks for your interest Goldemerald. A couple of comments:

- We do use DL. I didn't mention it because it is not relevant in the context of the post;

- How we use DL. One of the jobs of the graph is to tell the DL algorithm what content needs what type of descriptors. The graph can do this thanks to the different levels in our ontology, and because it understands our content in its context: it understands people's interaction and tagging. DL is a small part in our entire infrastructure;

- The DL market. Lots of companies use DL to identify attributes in an image, and the level they achieve is impressive. Having had long discussions with the best of these companies, I can tell you that building the correct ontology is nearly impossible without the entire infrastructure. We are happy building the intelligence that tells DL what to do, and then attaches descriptors correctly to outfits and taste profiles;

- Our patents. They cover a few relevant aspects in the online fashion market: a system to tag fashion images with shoppable products; a system and method to capture/understand how people mix and match clothes in outfits and closets; and more.


That seems incredibly tedious. For once I find myself agreeing with Zuckerberg about clothing choice.




Applications are open for YC Winter 2020

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: