From Paris to Berlin: Discovering Fashion Style Influences Around the World

Abstract

The evolution of clothing styles and their migration across the world is intriguing, yet difficult to describe quantitatively. We propose to discover and quantify fashion influences from everyday images of people wearing clothes. We introduce an approach that detects which cities influence which other cities in terms of propagating their styles. We then leverage the discovered influence patterns to inform a forecasting model that predicts the popularity of any given style at any given city into the future. Demonstrating our idea with GeoStyle-a large-scale dataset of 7.7M images covering 44 major world cities, we present the discovered influence relationships, revealing how cities exert and receive fashion influence for an array of 50 observed visual styles. Furthermore, the proposed forecasting model achieves state-of-the-art results for a challenging style forecasting task, showing the advantage of grounding visual style evolution both spatially and temporally.

Leveraging Multiple Relations for Fashion Trend Forecasting Based on Social Media

“…To investigate further on the external knowledge incorporation, we propose to leverage two types of external relations among fashion trends from two perspectives, the fashion element relations, which have been adopted by KERN, and user group relations. Al-Halah et al [11] explore fashion influences between different cities, the conclusions of which coincide and support our hypothesis to a certain extent that the fashion trends of different user groups are not independent as well. Specifically in our task, it is natural to consider that the trends of any fashion element for a user group (e.g., New York Female) are probably correlated with the corresponding trends of its subset (affiliated) group (e.g., New York Female 25-40).…”

“…One recent state-of-the-art study of fashion trends based on social media [2] extracted fashion attributes from social media images with a CNN model and investigated the trends of specific attributes in certain cities. Compared with other works which target on implicit fashion styles by visual clustering [10], [11], the fashion trends demonstrated in Mall's work were more specific. However, the fashion attributes it investigated to reveal specific fashion trends were coarsegrained and of less significance, sometimes showing seasonal patterns only, for example, wearing jacket.…”

“…Second, the e-commence data is not suitable for fashion trend analysis because the purchase decision is actually affected by many factors, and fashionability of the item is only one of them. Another very recent research [11] also adopted similar strategy. It trained a neural network model to detect the fashion attributes and learned a set of fashion styles based on attributes through Gaussian mixture model.…”

“…Such time series correlations have also been emphasized in other time series modeling tasks such as stock price prediction [31] and water and air quality monitoring [28]. In terms of fashion trend, various complex relations exist, for example, the lead-lag influential patterns across different big cities [11]. However, such complex relations are difficult to specify and explore.…”