4

Amazon's recently released its new Echo Look device, which is a modified Amazon Echo device with a camera added. One of the new features is a 'style check'. From TechCrunch:

The device also works with the company’s Style Check, a feature of the Echo Look app, which uses machine learning to compare different outfit choices, awarding them an overall style rating.

The app uses a combination of machine learning and advice from experts in the style space. Letting AI pick out your clothing in the morning should be a pretty interesting experiment.

I'm interested in how, exactly, the 'machine learning' determines which clothes are more or less stylish. I would imagine that style is a very subjective measure which is difficult to capture with a computer, but I'm not particularly familiar with developments with regards to using computer vision for fashion.

I'm not holding my hopes too high that Amazon would disclose how their algorithms work, but are there any similar approaches with regard to computer vision and style? Is there any information about how Amazon might have implemented this, or already-known methods?

Improve this question
2
  • As far as I know there is currently no more than "It is powered by both machine-learning technology and human opinion[.]" out there as explanation. Of course as a major fashion retailer Amazon does have a vast amount of pictures of models supposedly clothed in the latest fashion. That might give the machines a good start which then can be adapted by experts. Whatever an expert on fashion may be. However fashion is a very fuzzy subject so even if the algorithms fail we will rarely know.
    – Helmar ♦
    May 3 '17 at 22:01
  • Machine learning would be combined pattern recognization service and strong data crunch between what the person has now on and what fashion models for example have or those products that sell best in Amazon. How they do the calculations is some sort of data science OR if the recognition on both sides done well, a quite simple check if the product on people is on top 10.
    – mico
    May 4 '17 at 5:36
4

From Amazon publish:

"Submit two photos for a second opinion on which outfit looks best on you based on fit, color, styling, and current trends. Over time, these decisions get smarter through your feedback and input from our team of experienced fashion specialists."

So, they want to give their two best choices and from them machine will guess with wisdom the best.

This gives some hint what the machine should identify and learn to gain the wisdom:

  1. First of all identify which clothes in different pictures are same in real life. Helps to compare to other similar decisions.

  2. Identifying people characteristics to be combined to different type of clothing. Different clothes fit to different kind of people.

  3. Much own opinions of people collected from statistics of selling or the experts, maybe growling through fashion pages.

  4. What kind of clothes belong mostly to your own top 2.

  5. If there is some theory to be found what 'to fit' means to you and others.

  6. Etc

edit:

Pocket-lint opens the backgrounds and says the feature is originally a premium feature of Amazon's shopping app. I got an impression on the article that the analysis is maybe not quite instant and thus the wisdom behind maybe is only brute human source. Who knows.

Improve this answer

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.