DeepWear™ is a system that uses deep learning to capture the characteristics of various designs and generate new designs. While an outstanding designer can produce many fans, there is a limit to the amount of work a person can undertake. However, a computer incorporating a deep-learning neural network can learn by observing the products, and much like an outstanding assistant, can output many ideas for the new products. Moreover, by learning the features of many designers, it is also able to propose new brands by blending the features of multiple brands. In addition, it generates a variety of colors and designs that are similar to one another, making it possible, from the various options available, to choose a design that pleases the customers.
「DeepWear™」デザインの商品作成までのプロセス(例:スポーツグッズ)The designing process of a product using DeepWear™ (e.g., sports merchandise)
■ Case 1.
Case Study: Fashion Design
By feeding it a large number of fashion designs, we have developed a tool to assist designers and patternmakers in design creation. We have then confirmed that patternmakers can produce the clothes based on the designs generated by the AI. We are also working on the issues for the actual use, such as the outlines and the expression of details. We are also researching the functions to automate design analysis, pattern making, and material selection.
■Case 2.
Case Study: Sports Merchandise
Client: Hokkaido Nippon-Ham Fighters
By using DeepWear™, an AI-based design generation technology developed by PxDT, we have created designs for Hokkaido Nippon-Ham Fighters merchandise. Our second collaboration’s theme was “Athletes x Hydrangeas.” The designs were generated by combining the athletes’ play scenes with hydrangeas. Hydrangeas, which come into full bloom in Hokkaido in July, were generated to match each athletes’ personal color. The designs, which merge the hydrangeas and the athletes, were made possible because of DeepWear™.
■ Case 3.
Case Study: Wallpaper
Client: WhO
This is the wallpaper design that was selected from the 100 items automatically designed by DeepWear™, an AI-based design generation technology developed by PxDT, which was trained by feeding it with about 1,000 existing wallpaper patterns.
As with most of the existing wallpapers, the designs fed into the system included repeating patterns. Thus, if at the first glance the generated designs may also appear to be comprised of repeating patterns, a closer look reveals that the patterns are highly random in the details, showing fluctuations and differences in line thickness. While several generated designs may resemble the existing ones, this one shows a complex combination of learned individual features, a unique AI design that is difficult to reproduce by hand.
Learning the characteristics of a brand and generating new design proposals from multiple images (fashion design).
In the typical manufacturing process, first, the designer creates a 2D sketch of the design, which is then drafted into patterns by a patternmaker. The production worker will then use the patterns to cut the cloth pieces and then assemble them to complete the clothing.
Our system deep-learns the characteristics of a particular brand, and by extension the characteristics of the designers’ products, to automate the designing of the sketches used to draft the patterns. Also, in order to generate the images at the resolution needed by the patternmakers, we make use of Generative Adversarial Networks (GAN).
計算機・デザイナー・顧客の知能を融合させ、新たな服飾⽂化を構築
Building a new fashion culture by fusing the intelligence of computers, designers, and customers
We believe that we can build a new clothing culture by concurrently conducting research regarding the process of having users evaluate the finished products, as in the case of a GAN system. Moreover, it is possible to apply this to a service dedicated to direct ordering of tailored brand products that match personal preferences, by developing software through which the customers can by themselves enable latent variable adjustments.