Today marks the official start of New York Fashion Week and the Spring 2018 season. Over the next five weeks, here in New York and in London, Milan and Paris, close to 500 designers will show their collections to the world.
Digitalization has completely changed fashion week, to the point where it’s having an identity crisis. Thousands of runway images will flood online and fill our feeds. We consume them fast, and move beyond them even faster.
With so much speed, it’s hard to know what’s relevant. What will have a commercial impact, and what is truly new?
Over the last 10 years, the changes in the fashion industry have been stark. To tackle them, we built software to reveal which products sell (and which don’t) and how they’re priced.
We put in place the world’s first team of data analysts reviewing fashion weeks. Each season, these industry experts review 14,000 runway images to determine the key themes and trends, and highlight retail opportunities.
And over eight years ago, we developed color science to automatically quantify color trends on the runway.
But now our data scientists are tackling new ground. We’re using AI to make sense of everything that’s going on across these shows.
This is your crash course in AI and what it means for fashion.
What is Artificial Intelligence?
Artificial Intelligence sounds scarier than it actually is. Basically, it’s teaching a machine or piece of software how to identify nuanced and complicated things with human-like awareness, and then react accordingly.
It’s super buzzy in tech right now. Mostly because the focus over the last seven or eight years has been on collecting huge amounts of data. Now, this technology gets beneath those databases and drives insight. So how do data scientists get to AI?
First stop: Machine learning
If AI is the overall notion of computers understanding things, machine learning is an approach to achieving that. It’s a model where computers learn from their actions and from their mistakes, continuously improving their performance.
Go a level deeper and you have, well…
…Deep learning
Deep learning is type of machine learning. It’s a group of models that focus on understanding the features and representations within data in an attempt to generalise better. Deep learning tackles more complicated, less binary stuff.
While so much of the data industry focused on algorithms, deep learning adds extra dimensions.
An algorithm is like your friend that’s really into trains: great at something hyper-specific. But deep learning is more of an all-rounder: your pal who learns new languages with minimal effort because they “just get words” or the active friend who can pick up any kind of racket and be the best on court.
As far as data science goes, deep learning is the golden child. It’s newer because it’s only been made possible with technological advances in computing power. Many believe it’s the future of AI because the models behave more like the human brain, spotting patterns and subtle features in stimulus.
What can deep learning and artificial intelligence bring to fashion week?
With deep learning, data scientists are finding ways to make data incredibly insightful, and to build software that automates some of the decision making to help humans do their jobs better and faster.
That’s why retail is becoming interested. Thankfully, we’ve had a team of data scientists tackling these issues for years. And their most recent research is focused on fashion weeks.
Taking deep learning to fashion week
Our team has been building deep learning models that not only process thousands of runway images, but actually see, and give value to, components within them.
They do this by encoding the space within a runway image, attributing a reference point to everything the machine sees, whether it is color, texture, print or form.
When those references are lifted from the original image, they can be plotted elsewhere, they can be used to create a composite image, or they can be combined with other images to spot trends.
This stuff is incredibly exciting, because for the first time, we can get machines to spot the similarity and differences between collections. Software can actually translate how similar two collections are to one another, whether tracking one designer over several seasons, or two designers within the one fashion week.
Why is that useful? For one, if a machine can distill what constitutes a brand’s identity, we can layer in commercial data to understand what it is about a brand that makes people buy over time. We can quantify how far from that identity a new collection is and detect which parts will resonate best with consumers. That’s incredibly powerful for a buyer.
Creating something completely new
When machines lift those reference points from an image and rebuild them, they create a composite image, like below.
These garments never appeared in a show. They look similar to original runway photos, but they’re built entirely from code. The brilliance of this is what we can do next.
We can take a whole set of images’ codes, and find the mean. That means we can take an entire season’s worth of runway images, and create one image to summarise the look. That one garment didn’t exist, it’s made up of all the important parts of the whole set. That’s something not even the human brain can do.
Imagine a fashion critic being asked to summarise New York Fashion Week. They’d have to go through every image from every collection, taking notes and making personal biases and assumptions along the way. Within seconds, a machine could generate five composite silhouettes, colors and prints which distill the essence of a whole city’s designers. No bias, no preference, just fact.
Artificial intelligence summarizes fashion week at the click of a button. An editor would need days.
Machines can also be trained to look for similar things within a set of images. They can match similar shapes, colors or prints. This is another really tangible way of quantifying a new season’s trends. A buyer could instantly see exactly which designers backed the new-season longline trench, or which collections to look at if they want to stock metallics.
What’s next?
This fashion week, we’ll be testing out some of our new developments and having a little fun. We’re creating a challenge for you. We’ll use our deep learning tools to merge various collections, and create one composite image that acts as a midpoint between the two.
At the end of each fashion week, we’ll launch a quiz to see whether you know as much about runway fashion as our machines do. They’re firing up right now, so you’d better get revising!
These advancements that started as experiments are huge for this industry. They’re undoubtably going to change the way we understand fashion. All of us at EDITED are incredibly excited about the next frontier in data science.
Katie Smith
