The fashion industry isperhaps more than any other industry, characterized by an economic model based on excellent customer experience, product exclusivity and personalized service.

This means that brands are constantly searching for the latest trends, and analysing consumer behaviour and needs in order to exceed and even anticipate them.

Though the industry might not always have been at the cutting edge of digital technology, there is no denying that it is today increasingly at the forefront of the digital world, particularly in terms of social media. Instagram, for example, has gradually become an extraordinary hotbed for trends and influence in the fashion world. All of this has naturally led the fashion world to seek out technology capable of tracking and identifying these trends from the huge mass of data at its disposal. That's where artificial intelligence comes in, and, in particular, computer vision and machine learning.

Tell me who you are and I'll tell you what to wear

"Alexa, which outfit suits me best?" Launched in the US in June, Amazon’s smart camera Echo Look makes it easier for you to choose between a number of outfits. Using photos and videos, this "style assistant" analyses your look and recommends one outfit over another, based on the advice of fashion experts and the latest trends. Behind this small 15 cm device lies a whole host of visual technology and machine learning algorithms.

It is just one of the many examples of the incredible advantages artificial intelligence and machine learning have to offer the fashion industry today. To improve their customers' experience and create an increasingly personalized and bespoke service, brands can now count on increasingly high-performance mathematical models capably of analysing images, identifying patterns, determining trends and guiding customers to the product most likely to please them.

Image recognition technology has notably improved the features of visual search. The French startup Watiz offers a mobile solution based on computer vision and deep learning which, using a screenshot of a look, suggests a catalogue of similar outfits from a selection of online stores. The same service is offered by Wide Eyes Technologies,and the highly publicized Screenshop.

Computer vision is also playing an increasingly important role in the creative process, from the design of a collection to the prediction of trends. This is the subject of the Reimagine Retail project, led jointly by Tommy Hilfiger, IBM Watson and The Fashion Institute of Technology (FIT). Computer vision, combined with natural language understanding technology and deep learning algorithms, has proven its ability to produce key data on trends, making it possible to better anticipate demand and accelerate the design process.

In terms of predicting trends, the Frenchies at Heuritech are paving the way. Winner of the LVMH Innovation Award in 2017, their deep learning solution analysis pictures from fashion influencers on social media (types of clothing and accessories, colours, patterns, brands, etc.) to identify emerging trends. The goal: to give brands a head-start on their market, enabling them to be the first to identify local and international fashion trends.

How can algorithms predict trends that are, by their very nature, ethereal?

In a world that is, by its very nature, ethereal, in which trends come and go with the moods and tastes of Instagramers, lthe key is the relevance and evolutivity of the deep learning algorithms used.. Although computer vision technology has become more powerful and precise in terms of image detection, it is in fact the performance of algorithms that today makes it possible to analyse millions of images and connect them with external parameters to identify trends.

This requires advance classification of inputs (images) and outputs (what you want to detect) through meticulous preparation of the data sets used to train algorithms. It is not enough to simply identify a handbag. What really makes a difference is being able to identify a handbag from such-and-such a brand, in a particular colour, with a particular accessory, used in a particular context.

In other words, being able to take into account the diversity of images (format, context, resolution, etc.), the variability of items (brands, models, colours, etc.) and their distortion in the image (a handbag worn over the shoulder, placed on the floor or a chair, etc.), all whilst continually developing these parameters to ensure that the inevitable evolution of behaviours and tastes are taken into account. . This is why it is essential to use specific training data, refined according to the brand, item and trend you want to detect.

All the big-name brands like Dior, LVMH and Zara have taken at least a passing interest in AI, and in most cases they’ve taken a resolutely keen interest. It is also on the rise in the cosmetics sector, with algorithms now capable of recognizing the unique characteristics of a face (skin tone, wrinkles, red patches, etc.), making it possible for consumers to test makeup directly on their smartphone, for example.

Artificial intelligence has a promising future in the fashion industry. The possibilities seem endless. However, as cultural trends, social perception and subjective criteria are involved, if you want to get genuinely useful results, humans still need to prepare the data that will train your algorithms.

Originally published on July 26, 2018 Topics: Machine Learning Deep Learning Retail Computer vision

Subscribe

You may also like: