March 14, 2017 | Written by: Vikas Raykar
Categorized: Cognitive Computing
Share this post:
Next fall when you’re out shopping for the hottest fashion trends, you may come across a lot of gray, brown and black tones like nickel, umber, and raisin black, and for pops of color, sea blue and maroon. Fall 2017’s dominant color palette is just one of the insights Watson unveiled from analyzing hundreds of New York Fashion Week runway images from 12 designers in a first-of-its-kind partnership with global “fashion bible,” Women’s Wear Daily.
Visual analysis of each runway photo includes face, body, and pose detection, as well as color identification – all within about 1 second per image. (Photo credit: Women’s Wear Daily)
Fashion designers and fashion houses usually start conceptualizing and designing products for the new season six months to one year prior to the actual selling season–though in recent times this has been drastically reduced with the emergence of fast-fashion retailers. That’s why for most apparel retailers, and the fashion industry in general, knowing the trends customers would like to wear next season is extremely important. To achieve this, it’s important for apparel retailers to have the ability to analyze current fashion trends and also forecast future fashion fads.
From images to insights
Traditionally, fashion houses send fashion experts to the big four major runway shows who actually observe what is being shown and form a subjective opinion of the emerging mega trends. While there is great value in expert fashion observations, they are invariably limited by how many shows they can see. It is seemingly impossible for one person to manually look towards say 50,000 thousand images coming out from the runway shows. With a more data-driven approach, Watson can crunch all the images and then provide trend reports to the observers who can now use data based insights to back up their gut instincts.
For each runway image using a combination of face and body pose detection and other image segmentation algorithms we were able to determine the dominant color palette for the apparel in just one second per image. We then analyzed the entire collection of fashion images for 2017 to determine the dominant color palette for the season. This ultimately provides an entire runway show worth of insights in less than an hour. In a day, we might deliver insights on all of the images from every runway show in the world.
We were also able to analyze the similarities between various designers and how they are influenced by each other. We computed similarity score for each pair of designers, by aggregating the similarity between their image collections.
The seller (and buyer) benefit
A trend refers to a general direction (typically upward) in which something is developing or changing. A trend can arise from a multitude of sources. For example, a trend toward animal prints could mean that high-end fashion designers have started showing animal print designs on the runway, apparel retailers have started to introduce them in their online catalogs/stores and are quickly selling out, celebrities have been spotted sporting animal prints, fashion magazines, websites, blogs and social media sites have started recognizing this trend, and many fashion-forward consumers have starting wearing animal prints on the street. At IBM Research, we are building a cognitive fashion agent that can analyze current fashion trends from multiple sources (catalogs, articles, blogs, images, social media) and forecast future fashion trends.
This analysis brings great value to fashion merchandisers who can very quickly and decidedly get a feel for what is being shown at the major runway shows in order to make an informed decision on inventory. Here are some examples of what industry experts might do with this information:
- Designers and product developers can understand what are the macro trends that are emerging.
- Local fashion houses may adapt trending runway looks to make them more palatable for local audiences.
- An e-commerce site may decide to show the most trending colors/items on the first page.
- An e-commerce site can give suggestions on what items and their attributes (like color) go well with each other based on such trend reports.
- A consumer could search on an e-commerce site with a simple query like ‘show me some dresses in trending colors.’
The industry’s speediest intern
Fashion is a highly creative industry and computers are definitely not creative enough to replace designers. Our goal is to provide data-driven insights that give designers the tools so they can design better. For example, a designer cannot possibly look at a collection of 500,000 fashion images from the last decade – this is where our cognitive agent can help summarize them and provide novel ways of exploring these fashion images to inspire better and more informed design.
Similarly, pouring through literature and images to make informed decisions on a day-to-day basis can be done efficiently with the help of a smart machine.
For more information on the IBM Research cognitive fashion technology, visit the IBM Marketplace or contact Jaikrishnan Hari.
The insights for this partnership were provided by the Cognitive Fashion team at IBM Research, India (Vikas Raykar, Raghav Singh, Ayushi Dalmia, Priyanka Agrawal, Mohak Suhkwani, Sachindra Joshi).