Scenery-Based Fashion Recommendation with Cross-Domain Geneartive Adverserial Networks
Sang-Young Jo, Sun-Hye Jang, Hee-Eun Cho, Jin-Woo Jeong · 2019
To build an effective fashion recommendation system is a still challenging issue due to its high complexity. Previous research works generally have focused on how to provide fashion items visually similar to the user's current fashion taste. However, a scenery (natural landscape) around users is also an important affective factor in recommending fashions. This paper presents a novel system to recommend fashion designs that fit target sceneries. To address this, the exemplar photos regarding the target landscape are first collected from the database. Afterwards, a cross-domain generative adversarial network (GAN) is applied to generate fashion designs from the sceneries. The experimental results demonstrate the feasibility of the proposed system and imply further research directions.