98 8. RESEARCH FRONTIERS
(a) Datasets. (b) Codes.
Figure 8.1: Illustration of the website pages.
8.1 GENERATIVE COMPATIBILITY MODELING
Currently, the generative adversarial networks (GAN) have achieved great success in image gen-
eration tasks, such as image retrieval, video generation, and image synthesis, which propels the
attempt of generative compatibility modeling. In the context of recommending a bottom for the
given top to make a compatible outfit, if we can first generate a bottom template as an auxil-
iary link to the top, then the compatibility can be further modeled between the template and
the given fashion item, which may improve the recommendation effect with the item-template
compatibility modeling.
In light of this, the generative compatibility modeling aims to measure the compatibility
between fashion items with the auxiliary complementary template generation. To the best of our
knowledge, limited research studies have been devoted to the generative compatibility modeling,
possibly due to the following challenges: (1) to accurately measure the compatible preference
between fashion items, the latent representations are adopted for the item-item compatibility
modeling. Furthermore, the compatible template for the given fashion item may contribute to
the auxiliary item-template compatibility. us, the main challenge lies in how to seamlessly
..................Content has been hidden....................
You can't read the all page of ebook, please click here login for view all page.