To build a predictive model, we identify two key properties of an influencer-time precedence and novelty-and use a statistical measure that captures these properties to calculate the degree of influence between cities or brands as well as the influences among the various styles themselves. Propagation of inter-brand and inter-city influences. Our results indicate the advantage of grounding visual style evolutionīoth spatially and temporally, and for the first time, they quantify the Model achieves state-of-the-art results for challenging style forecasting The discovered influence relationships reveal howīoth cities and brands exert and receive fashion influence for an array of Styles move between locations and how certain brands affect each other'sĭesigns in a predictable way. Our model learns directly from the image data how Photos from 44 major world cities (where styles are worn with variableįrequency) as well as 41K Amazon product photos (where styles are purchased Toĭemonstrate our idea, we leverage public large-scale datasets of 7.7M Instagram The future popularity of any given style within any given city or brand. We then leverage theĭiscovered influence patterns to inform a novel forecasting model that predicts Other entities in terms of propagating their styles. We introduce an approach that detects which of these entities influence which WeĮxplore fashion influence along two channels: geolocation and fashion brands. We propose to discoverĪnd quantify fashion influences from catalog and social media photos. Intriguing, yet difficult to describe quantitatively. The evolution of clothing styles and their migration across the world is