Yudai HONMA (The University of Tokyo) http://www.honma-lab.iis.u-tokyo.ac.jp
Fashion press (Carlin, Inc.) https://www.fashion-press.net/
2. Points of the presentation:
- We built a system which can arrange the format of a total of 160 Men's & Women's fashion collection imagery, including Louis Vuitton, Gucci, Prada etc. (approximately 7000 images per season), making them into a database in order to mathematically analyze the latest fashion trends.
- A cross-search function on mix and match outfits proposed by any major high fashion brands, focusing on color and item, is published on the Fashion press web site, so that everyone is able to design an outfit of their own taste, regardless of brands and or collection seasons.
- By using this system efficiently to prepare many feature articles of the trends of 2019 Spring & Summer collections, we've made it possible to promptly publish articles with a new viewpoint, which holds both the unique sensibilities of our fashion editors, as well as the perspectives of the objective analysis results.
The research group lead by Associate Professor Yudai Honma of the Institute of Industrial Science, the University of Tokyo and, Carlin, Inc., who runs the on-line media, Fashion press (Note 1) have developed a unified data managing system of the fashion brand collections as a mathematical analysis data, and has built together, a system that can analyze and evaluate the latest fashion trends.
We have published a new gallery, where people can conduct a unique cross-search by color and item, out of the huge collection imagery (images of the fashion collections Note 2) data, held by Fashion press.
There are no other examples of a cross-search function on mix and match outfits proposed by any major high fashion brands, and we believe that general users will be enabled to pursue fashion according to individual preference. In addition, Fashion press will be publishing many articles featuring the 2019 Spring & Summer collection, written by our editors using the mathematical analysis system, aiming to further appeal and expand our digital contents.
<Background of our research>
The majority of the high fashion brands, including Louis Vuitton, Gucci and Prada, display their latest collections in runway fashion shows twice a year in cities such as Paris, London, Milan and Tokyo. These collections, filled with the designers' creativities, are significantly vital to capture the season's latest trend. Fashion press holds over 30,000 images of approximately 800 collections every season, which is one of the largest number of such images held by a Japanese company, and we publish them on our web site in a gallery format for our users to enjoy.
On the other hand, because so many collections are introduced every year, it had been a challenge for us to specify which imagery the users actually want to browse, with only brand names and collection seasons to narrow the choices down. We needed to create and utilize a suitable database for the huge amount of our collection imagery data.
<Outcome of our research>
For the purpose of being able to handle a broader amount of data held by Fashion press more objectively, and also to obtain as much information from the images as possible, The Honma research group of the Institute of Industrial Science, the University of Tokyo, and Fashion press have been researching how we could analyze the collections data using a mathematical approach. The themes we considered especially important were, (i) the establishment of a unified data managing system, and (ii) the construction of a mathematical analysis method:
(i) Unified data management of the collection imagery by a suitable format for mathematical analysis
Each collection image is an image data, which is not a suitable format for mathematical analysis. Therefore, we arranged the format of our 2014 and on data, in order to facilitate data handling and at the same time, to unify the management of the huge amount of data we had. Due to the establishment of this unified data management system, we have been able to implement mathematical analysis on the collection imagery much easier.
(ii) Construction of a mathematical analysis of the collection imagery, and it's implementation as a new system
Next, we examined the various options of mathematical analysis, which may be made of these data. As a result, we confirmed its validity regarding cases such as statistical evaluation of a biased data, and clustering (Note 3) due to the similarities of some of the images. Therefore, we have decided to implement this mathematical analysis method as our system, which our editors would easily be able to use.
From Friday, February 22, Fashion press web site will be publishing various new contents including (A) the new gallery, and (B) feature articles.
(A) The new gallery, with a cross-search function using color and item
The huge collection imagery data, which was only searchable by season or brand, is going to be renewed and will be able to cross-search by color, item, or by combination of the two. The users will not be bound by brand names, and will have the freedom to choose from color and or items to search their own unique fashion outfits. Unlike similar services targeting reader models, our service targets collections by high brands, enabling general users to independently pursuit a further elegant fashion coordinate.
(B) Feature articles of 2019 Spring & Summer collection, using the mathematical analysis system
Our editors are able to use the mathematical analysis system we have constructed in this research, in order to perceive the sort of data held by Fashion press, in a multilateral and objective way. This will enable us to efficiently publish feature articles from a new viewpoint, focusing on the trends of the upcoming 2019 Spring & Summer collection.
It is our mission as a company operating a portal site, to arrange information in an appropriate form for the needs of our users. In this research, we aimed to create value out of such restructuring of information, and we believe that this could eventually lead us to a new fusion of design and engineering.
Institute of Industrial Science, the University of Tokyo
Associate Professor, Yudai HONMA
Fashion press Editorial Department（Carlin, Inc.）
E-mail：support (please append '@fashion-press.net' to complete the email address)
Note 1) Fashion Press
An on-line lifestyle magazine, focusing mainly on fashion as well as beauty, art, gourmet, music, movies, and more. Based on web technology, we offer the latest concepts and ideas of lifestyle, to a growing number of users. Organized in 2009. 52,000,000 PVs/month, 6,000,000 unique users.
Note 2) (Fashion) collection, images/imagery
Collections refer to Fashion weeks, where High Fashion brand designers introduce their latest fashion in runway shows, presentations, and or showrooms, which take place in Paris・France, Milan・Italy, New York・U.S.A., London, U.K. The Paris, Milan, New York, and London fashion weeks are referred to as the "Big Four", and these cities hold a fashion week twice a year for the Spring & Summer and Fall & Winter collections, both for men's and women's at different timings. In addition to these four major cities, fashion weeks are held worldwide, in various cities such as Tokyo, Shang-hi, Sydney and more.
Images/imagery is something like a catalogue, which is made by each fashion brand every season. The format may vary according to the brand; some may be a collection of photo shots of models wearing the latest fashion, and some may be with photo shots of runway shows.。
Note 3) Clustering
A mathematical method for dividing a large number of data into several small groups with similar trends. By focusing on the difference of the numbers of each group data, it is possible to find out certain tendencies of each group.
Fig. 1: Example of cross-search results by color & item