[2K5-GS-2] Machine learning

[2K5-GS-2-02] Trend Analysis-Based Fashion Image Generation

〇Ryozo Andre Leon Masukawa¹, Sayaka Aoba¹, Yugen Sato¹, Shosuke Haji¹, Makoto Sato¹, Taiga Matsui², Keita Ishikawa², Tomohiro Takagi¹ (1. Meiji University, 2. airCloset, Inc.)

Keywords:Generative Models, Fashion Trend Analysis, Image Retrieval, CLIP

One of the most essential tasks in fashion retail business is to predict the inventory by analyzing the future fashion trend. Generally speaking, fashion designers design a novel clothing by converting iconic designs shown in catwalk show into a general design that consumers use in their daily lives. As a result, fashion experts have a duty to detect novel fashion trend by aggregating external information sources such as fashion web magazines. However, external fashion information sources are too diverse for experts to analyze all of them objectively, constantly, and equally. To deal with this problem, we propose a system that generates images of novel clothing designs that matches existing fashion trends by combining generative models and information retrieval models. Experiments show that our proposed system can generate high resolution clothing images that reflects iconic fashion trend information.