Do Androids dream of Prada? How AI is already affecting the world of clothing
The fashion industry today is estimated at $ 3 trillion per year, or 2% of world GDP (according to the McKinsey Global Fashion index). Moreover: it grows annually, an average of 5.5% every year in the last ten years. Nevertheless, many giant brands have a hard time. Relying on several key figures involved in the design and prioritization of growth, they lost. New-old Victoria's Secret boss already lost his company more than $ 10 billion. Profit H & M falls already ten quarters in a row. Even the most successful brands, like Tommy Hilfiger and Michael Kors, are afraid to be in their situation.
Various technological start-ups and their own newly-formed IT-departments of the “fashionable” companies themselves are taken to solve the problem. Using the technology of machine learning, computer vision, deep learning, natural language processing and data mining, as well as feeding huge amounts of information about the fashion industry to systems, they try to predict what will be in trend in a few months and get at least some competitive advantage. The human element in this seemingly most creative industry may soon not remain at all. 3r318.
A simple solution to complex problems
In London, the last three years is going to 3-33-330. conference 3r3270. KDD, dedicated to the integration of neural networks and intelligent algorithms in the fashion industry. It discusses more than thirty topics: how to analyze the data, how to determine whether your robot has something to do with it, how to teach it to choose the right textiles, what system to use to assess the “fashion” of a product
This conference is full of participants: many large brands have been holding their own AI department for several years, realizing that the costs here are relatively small, and the potential profit can be incredible. One unclaimed line of clothing that goes to scrap can cost hundreds of millions of dollars. A clever robot, able to quickly understand the petabytes of information and identify unobvious connections, may suggest that launching such a collection now would be a bad idea.
Although - this is still a dream. Even H & M, which in this case advanced further than others, so far they do not trust their algorithms for 100%. But there are things that are already working, and show results. One of the most important contributions of AI to the industry was the evolution of visual search (for example, Unbxd.com ). With its help, online retailers can much more effectively predict what things the buyer would actually like to receive in his issue, based on any available data about the client.
Hundreds of companies are starting to use this technology (small - with Unbxd, large - with a proprietary system). The number of clicks that a customer has to make in order to get to the product he needs and make an order is reduced several times. The competitive advantage is undeniable: the client is much more inclined to buy clothes on the site, which can quickly offer him something interesting.
3r3391. How the robot decides what suits you [/i]
Buyers can now also simply take a picture of jeans or sneakers that they are interested in and get online matching of this photo. This functionality is already offered by the Amazon application, it is in Google’s smartphones, and Instagram is going to offer a similar solution. "Trendy" companies are counting on it. If everything is clear with gadgets and technology, the name of the model is often engraved directly on the case, then the particular name of the garment is often not even known by the person who bought it. With the help of an AI analyzer, getting “oh, exactly the same blouse!” Becomes easy: take a photo - clicked “Order”. And the profit of manufacturers is growing.
But this is the most primitive model. Smart algorithms are already capable of more. Several companies are already experimenting with data exported from your calendar. Their products find information about your future journey or (using cookies) about the weather in your location. And then on the sites of the stores recommend you current clothes or shoes. Say, if you are going to Sochi soon, you will suddenly begin to offer T-shirts. If you are going to Norway, a wide offer of sweaters is taken from nowhere. It will rain the day after tomorrow - they recommend that you take a jacket and waterproof boots.
According to The Cut, people now use on average only 20% of their wardrobe. We find what we like the most, and then we wear it all the time, and the rest of the things, which are “not so much our style”, simply gather dust in the closet (by the way, it's funny: 25% of clothes in the developed world are not used because that people are waiting to lose weight in order to fit in it). AI can help us use the potential of our purchased clothes at 100%. Pureple , for example, offers things that will go well with the clothes you already have, creating a full-fledged outfit. The company's own database and user reactions are taken into account. Do not google "whether these jeans come with a plain white T-shirt." Based on the decisions of previous users, the program will decide for itself what looks good together and what doesn't.
So far, however, Pureple and other similar applications to it are limited in the most basic - enough information that can be analyzed. Other companies are not eager to share their data. As a result, Pureple is not yet particularly effective and useful - which means that it does not have enough active users who, by attaching their favorite outfits, as in Tinder, would help the algorithms to gradually improve.
3r3391. Some designs generated by AI based on prototypes [/i]
Therefore, the main progress in creating a steep "fashionable" AI all are now expected from the largest companies. They have enough users and information about their real preferences. The two main fighters in the ring are Amazon and Facebook. Both giants have been doing their own AI designs for several years now. Facebook is preparing to turn Instagram into a marketplace, and Amazon is designing algorithms for its own brands, Ella Moon for women and Buttoned Down from 3r3102. Goodthreads
AI, developed by the Amazon team, is already able, "having absorbed" information about a certain fashion style, from scratch to generate new items in the same style. The task of the designer ultimately comes down to choosing the best options from these hundreds of generated garments. Moreover, the smart system Amazon gives a greater coefficient to items with better statistics on views and purchases on the store's website, so that new things have a better chance of becoming in demand. Where previously a command was required, now one algorithm works fine.
Here are three more promising directions from this sphere, and three applications that are already solving real problems: 3r3-33271.
Finery - the personal AI-stylist
takes your clothes, reads any new items purchased from the connected e-mail, and selects for you "bows", that is, combinations that look best on you. Plus, she recommends new things, considering what you may lack. The developers claim that their machine even takes into account the user's style. That is, if you have some caps and lowered jeans, she realizes that you should not recommend a tie or Fedor. Instead, AI will look at something “on the same topic,” like some kind of hoodie.
Amazon also launched a similar platform specifically for its website this summer. He has a “smart assistant-stylist” Echo Look . This is a camera with a built-in assistant, Alex, who stands in the apartment, takes a picture of you and estimates what you are wearing. Of course, they do not give you a rating, and the device will not say “you are terrible, 2 out of 10”. Instead, it allows you to compare two outfits, and then says which one she likes more, and how much (as a percentage).
And, of course, for the sake of which everything was intended. The application in the smartphone that the camera works on, then takes all these photos and, based on them, recommends other items of clothing for you to buy from Amazon, and also lures you to subscribe to all sorts of “fashionable” magazines. It uses the same Kindle model. Even selling the device at a loss, Amazon then plays on all fronts.
Hook - tracking trends for retailers
Search engine 3r3149. Hook
developed by Intelligence Node, designed for shops and brands. Everything is simple: if a person does not find what he needs, in the first ten points of issue, the probability that he will simply leave the site, 3r3151. rises
astronomically, a few dozen times. Therefore, it is necessary to quickly issue something relevant - to delay the buyer, so that, at least, the system will turn on, "well, if I killed so much time for it, it would be time to take something already."
To achieve this effect, Hook takes into account fashion trends, collects any possible data about the client and finds what to hook him. The deep learning algorithms on a saucer present to the retailer all the information that he can use against you. Your location The price level you usually focus on. Typical keywords that you score on search engines. Likes under pictures and posts on Facebook. As a result, the issue does not always show exactly what you scored, but, potentially, something more interesting. Let's say you wanted to buy a sweater, and here is warm golf, and even in your favorite color. So good luck! Exactly, you can take such a turtleneck!
Now the application collects information about products from ?000 brands, from H & M, JC Penny and Gap to Nike and Adidas. In addition to the search function, it is planned, like Finery, to add a recommendation service - so that the user can upload his photo, and the system gives him all possible recommendations so that he does not leave without a purchase.
Stylumia - prediction of the styles of the future
The program, which, according to the company developer, portends what will be in vogue in a few weeks or months. 3r3176. Stylumia
combines computer vision, natural language processing and data on people's behavior, which ultimately allows the system to predict fashion trends.
The application is designed for manufacturers and sellers of shoes or clothing. They enter, introduce colors /products /styles that they are interested in. Then the system, taking into account this request, collects for itself a huge array of relevant data. Information about the products from the site of the brand itself. Data on user engagement. Photos, videos and text markers in social networks (Facebook, YouTube, Instagram, Pinterest).
Further, Stylumia handles all this with AI and Deep Learning, and provides the client with squeezed useful information. Charts about what this product is generally thought of now, and what things you can expect from it in the future. What colors are in fashion, which are declining, which are growing. What kind of things are being discussed now. This helps the store or designer to make an informed decision: is it worth starting a similar project, will there be a demand for it, what of the similar is already on the market, what are the chances of success?
An example that Stylumia gives in blog. concerns fashion shows. At one big presentation, the 40 largest designers can show a total of 2?000 new items of clothing for the next season. No one person in themselves so much information will not hold. And here the program itself highlights all the most important things, and it looks at what has got accustomed, what they say, what conclusions can be drawn from this, whether this style will be popular next season, or whether it has already outlived itself
If you have ever read fashion magazines, the editors are constantly trying to figure out what color will be in the trend? What item of clothing is in demand? What combinations of colors and patterns now capture the minds? With a system like Stylumia, there is no reason to guess: computer vision immediately notes characteristic trends. For example, in the fall-2018 "in vogue" bright yellow and black on red. And the most popular individual thing was (at least in fashion shows) white women's blouses with puffy sleeves. That is, from the point of view of Stylumia, it was these items that retailers had to buy in order to maximize their sales.
The company currently has 40 customers - including Aeropostale, Puma, Myntra, Fastrack, Biba and Global Desi. They consult with the system to determine the appropriate designs for their new clothes. The company itself says that sales increase many times over, and collections collected “under the guidance of” Stylumia disappear from the shelves in just two or three weeks. The manufacturers themselves do not disclose any figures.
The last bastion
But can soulless algorithms really understand and replace fashion? Doesn't this need creative creativity and talent?
In fact, many experts believe that a good AI will cope with the task as well, and in many respects better. Michael Ferraro, executive director of the New York 3-333226. The Institute of Fashion and Technology
Fashion and so now largely fabricated. The fact that the next season will be "in trend" is decided by the boards of directors of large companies. Now just robots will take over this job. In my opinion, it will be even more honest.
Artificial intelligence technologies can also help to finally make the fashion industry more environmentally friendly. Now it is the second largest pollutant in the world (after the oil industry). The world sells 80 billion (!) Garments every year, using heaps of dyes and chemicals. And almost everything that was produced, pretty soon eventually ends up in landfills. Matthew Drinkwater, the head of the London agency Fashion Innovation, says: 3r-3271.
We produce and throw away too much clothing. In Britain alone, we get rid of more than 30?000 tons of clothing every year. If AI can help us produce only what is sold, the amount of waste can be seriously reduced.
The current model of retail does not work, we just produce garbage, it needs to be changed. Artificial intelligence will allow us to regulateproduction in real time, on the go adapt designs, taking into account the ever-changing demands of customers. Clothes created in this way can be more valuable for buyers, they will not throw it out so quickly. Who knows, maybe people will even fix their clothes. It would be great.
P. s. 3r36262. High-quality clothes American brands, from Timberland and Gap to Tommy Hilfiger and Michael Kors, can often cost less than goods from Russian clothing markets. Seriously. Especially if you take it directly to the US, and are not afraid to go to the Sale section (discounts are always real, with the exception of sites with clothes for children, oddly enough). And Pochtoy.com is able to deliver any purchases from the States to Russia. From $ ??? per pound.
Readers Habr after registration receive a discount of 7% for the HABR code.
It may be interesting
Thankyou for sharing the data which is beneficial for me and others likewise to see.Gulf Coast Western Reviews