Garment compatibility prediction method based on hypergraph

Abstract

The invention discloses a hypergraph-based clothing compatibility prediction method, which comprises the following steps of: constructing a hypergraph through an existing data set to represent suits and single clothing items in the existing data set, and dividing the data set into a training set, a verification set and a test set by utilizing a graph segmentation technology; then visual features are extracted through a convolutional neural network, and text features of the visual features are obtained; according to the method, a hypergraph is introduced to well reflect a complex relationship between a suit and a single clothing item, meanwhile, the hypergraph is converted into a simple graph to obtain a simple graph enhanced network model, a message propagation mechanism is utilized to iteratively update graph node representation in a simple graph structure, and final representation of graph nodes is obtained through gating circulation; the prediction capability of the model is enhanced through an attention mechanism, and the compatibility score of the suite is calculated by using the final representation of the graph nodes. According to the invention, the designed simple graph enhanced network model can well represent the relationship between the single clothes items in the suit, and can well predict the overall compatibility of the suit.

Description

technical field [0001] The invention belongs to the field of computer vision, in particular to the compatibility prediction in clothing matching based on neural network. Background technique [0002] In recent years, with the rapid development of the Internet and the growing maturity of the logistics industry, online shopping has become more and more important to modern consumers, which has greatly promoted the development of the fashion industry. According to statistics, the global online fashion market was worth $533 billion in 2018 and is expected to increase to $825 billion by 2022. Clothing is a daily consumer product, and consumers have a relatively large demand for it, with high consumption frequency and high density. In 2018, apparel accounted for 65% of the market, followed by footwear (25%) and bags and accessories (10%). Additionally, there are fashion community sites that allow customers to create their own styles using images of clothes on the site. However, ...

AI Technical Summary

Problems solved by technology

Their representation methods have certain defects, such as pair representation cannot reflect the complex relationship between multiple clothing items; sequence representation itself is unreasonable, because the relationship between clothing items in a suit is not orderly ; while the graph represents the modeling of the suit as a graph, but when building the relationship, it essentially builds the relationship between pairs

Therefore, the current research results have not found a reasonable representation method that can well represent the relationship between suits and clothing items. It is urgent to propose a new idea to better reflect the relationship between multiple items in a set of clothing. dense and complex relationship

Images