基于生成式对抗网络的用户下装搭配推荐

为解决消费者由于频繁购入相似服装以及不知如何穿搭的问题,设计了一款智能搭配系统,为用户提供穿搭建议,减少重复购入相似衣服导致的浪费。利用爬虫技术获取大量中高端品牌的服装搭配数据,利用深度学习的新兴模型生成式对抗网络,对服装搭配数据进行学习,挖掘搭配的颜色、款式等视觉规律,训练模型能够实现输入上装图像时智能生成下装图像功能,再通过图像相似度计算匹配到用户预设的个人衣柜,最后结合温度为用户推荐合适的下装。通过对比原搭配和模型生成搭配,验证了该方法的有效性,为智能穿搭提供了新思路。

Recommendations for user's bottoms matching based on generative adversarial networks

In order to avoid repetitive purchases of similar clothes and to solve the clothing matching problem, this research worked on an intelligent clothing matching system for providing recommendations to customers and end-users. A crawler technology was used to obtain a large number of clothing matching data from mid-to-high end brands, and the model of deep learning-generative adversarial network (GAN) was adopted to learn clothing data to explore the visual perceptions of colors, styles, and so on. When inputting a piece of top clothing in the system, a bottom picture can be generated intelligently and then matched to the user's personal wardrobe together with the consideration of the surrounding temperature. The effectiveness of this method was verified by comparing the original matching with the model generated matching.This work provides new ideas for intelligent clothing matching.