The fashion industry is often gridlocked in a paradox that’s rooted in the tension between its creative side and the bottom line. And in a retail environment undergoing extreme upheaval due to the demands of consumer centricity, this tension is being further strained.
But new technologies may be able to help — and some have knowledge greater than our own. Take Watson, IBM’s new artificial intelligence solution with the ability to crunch more data faster than anything else.
So, WWD wondered, what would Watson say about the latest collections shown during New York Fashion Week? Would it detect things merely human buyers and fashion editors missed? Would it at last be able to answer the question that the industry has pondered for decades: Is there a universal fashion truth?
That may be even beyond Watson. Nonetheless, WWD asked IBM to discern key colors, trends and even comparisons among New York Fashion Week fall collections that resulted in a comprehensive report on its analysis.
Watson made a splash for its introduction to the market in 2011 — appearing on TV show “Jeopardy!” and defeating former champions. IBM scientists developed the software on a Deep QA platform that runs on 2,880 processors activated by Power 750 computers. The technology is able to hold information equal to that found in one million books.
The development of this technology marked a new phase for machine-learning possibilities. Not only is Watson capable of applying a human-sense of uncovering facts, it does so with precision and speed. In the case of its “Jeopardy!” upset, Watson used more than 100 algorithms to source the most accurate answer. Since then, IBM researchers and scientists have only continued to evolve the technology. “At IBM Research, we are building a cognitive agent that can analyze fashion trends from multiple sources such as catalogues, articles, blogs, images and social media and forecast future fashion trends,” said Vikas Raykar, a researcher at IBM.
The Fashion Test
Access to fashion shows and thus the ability to gain exclusive understanding of emerging trends was formerly reserved for a minute group of merchandisers, editors, fashion forecasters and other insider roles — Watson democratizes fashion analysis. In an effort to comprehend the broad expanse of lines shown during New York Fashion Week, WWD selected 12 designers, both emerging and established, to review. They were Coach, Jonathan Simkhai, Delpozo, Marc Jacobs, Alexander Wang, Brandon Maxwell, 3.1 Phillip Lim, Public School, Ralph Lauren, Prabal Gurung, Jason Wu and Dion Lee. Watson analyzed 467 runway images to determine the results.
IBM employed two main methods in the study. Its Research Experimental Service coined the phrase “cognitive fashion” to describe the suite of application programming interfaces, assets and use-cases tuned for the fashion industry. “Before we provide the aggregate trend analysis, we apply a series of what we call fashion annotators such as face, apparel and pose detection in addition to color attributes that takes one or two seconds per image to run these annotators,” said Raykar.
Watson also utilized its visual recognition tool to break down aspects of each look to tag the image, find human faces and find similar images in a collection. Within this, Watson utilized body and apparel detectors to discern both body and apparel details.
Additionally, IBM’s Research Cognitive Fashion asset — an app developed for the purpose of this report — determined the main colors in each runway image. “Dominant colors for a collection of images were obtained using color trends asset from IBM Research’s Cognitive Fashion [app],” the report said. What’s more, it also catalogued the similarity between images.
The experiment included the deployment of a visual browsing app capable of discerning similarities between designers and repeating trends. “This tool is useful to interactively explore all the images and also see how fashion designers get influenced by each other,” said the report. “The notion of similarity can cover various aspects like color, pattern, cut and silhouettes.”
Dominant colors: For each runway image, Watson extracted the dominant colors of the styles. This bank of colors was analyzed to mine the top shades for fall. Watson determined that the most popular palette included Pantone colors Raisin Black, Pastel Brown, Pale Silver, Cedar Chest and Deep Koamaru. These share commonalities of moody neutrals and rugged earth tones found in Pantone’s key 2017 color palettes, Grand Canyon and Ethereal Materials.
Repeating patterns: Watson also discovered similarities in prints and patterns. It found comparable motifs to those seen at Prabal Gurung and Jonathan Simkhai. This is profound as fast-fashion retailers, home interiors and CPG companies have often had to rely on trend analysis companies to divulge which trends are the most worthy of download and manipulation. Now with the use of this technology, the need for outsourcing these services is lessened.
The web: After analyzing the collections, Watson drew conclusive correlations and comparisons. “We wanted to analyze the similarities between various designers and how they are influenced by each other,” said the report. “We computed similarity scores for each pair of designers by aggregating the similarity between their image collections.”
The most in sync? Brandon Maxwell and Alexander Wang, which was somewhat surprising. Though both bright young things in the industry, Maxwell and Wang’s aesthetics usually differ. The former’s pieces encapsulate refined, elegant eveningwear frequently seen on red carpets while the latter’s streetwear-inspired, elevated-grunge pieces are gobbled up internationally. “Alexander Wang and Brandon Maxwell are the closest in terms of color, cuts and knee-length,” said the report produced by IBM, detailing the results.
All designers were included in a web to depict how similar they were to one another. “We plot a graph where nodes are the designers and the edge length represent the similarity scores between two designers,” the report said. “That is, the shorter the edge the more similar are the designers.”
Observed at NRF in January, artificial intelligence and machine-learning technology are seen as essential solutions to help retailers succeed in a challenging market. Upending the classic infrastructures of product developers, merchandisers and marketers, Watson is a disruptor in the purest sense. Shifting workflow from seasoned experts to computer scientists who needn’t speak fashion’s sometimes obtuse and deeply intellectual vocabulary poses both benefits and drawbacks.
Meanwhile, speed-to-market is nearing sprint levels of tempo. Today’s product life cycle process requires the ability to analyze trends and then develop designs relevant for a specific audience — and to do so at an accelerated pace. Keeping step with consumer demands is only possible with the use of the proper tools — product life-cycle management systems, AI and data collection and analysis. And these technologies require data scientists and engineers who are capable of culling the most information from it.
Perhaps Watson and other AI solutions can serve as a convergence technology between the data science side of the business and the creative side. But this convergence of fashion and technology will require discipline and balance. At this point, data scientists and designers will have to work in tandem. Designers will bring to the table context for the data to make sense, and to help inform product development — which also requires having insights into various audiences and consumer preferences.
From here though, retailers and brands can either train existing staff in technology such as Watson or educate computer scientists on what makes a fashion trend work for the masses — determining the answer to which is as much art as science.