Modeling Fashion Influence from Photos
Ziad Al-Halah Kristen Grauman
[Paper] [Data] [Code] [Results] [Related Projects]
We propose the first work to model fashion influence relations among fashion brands and major cities around the world learned from photos.
In this work, we build on our earlier findings on modeling fashion influence relations between major cities around the world. In particular, we extend our model to go beyond cities' influences to capture how styles themselves influence each other. We propose a model that leverages both cities' and styles' influence relations, and we show it yields higher forecasting accuracy. Furthermore, we provide an in-depth analysis of the relations discovered by our model and how the measured influence correlates with fashion experts and public opinion on what is fashionable. Finally, we generalize our model to capture not only geographic influences but also influences among fashion brands. We demonstrate that our approach captures how major fashion brands influence each other with a new set of experiments on another public dataset, AmazonBrands.
Paper
Modeling Fashion Influence from Photos
Ziad Al-Halah and Kristen Grauman
IEEE Transactions on Multimedia (TMM), 2020.
[paper]
[on IEEE TMM]
@article{al-halah2020b,
title={Modeling Fashion Influence from Photos},
author={Ziad Al-Halah and Kristen Grauman},
journal = {IEEE Transactions on Multimedia},
doi = {10.1109/TMM.2020.3037459},
year={2020}
}
Data
AmazonBrands
- The brand annotations for the dresses dataset can be found here: Dataset
- The learned style trends from the brands can be found here: AmazonBrands trends
The two types of learned trends for 50 styles in 44 cities from GeoStyle used in our work:
Examples of Discovered Fashion Influences
Influence relations among fashion brands
Style influence relations discovered by our model among fashion brands. The number of chords coming out of a node (i.e., a brand) is relative to the influence weight of that brand on the receiver. Chords are colored according to the source node color, i.e., the influencer. While some brands have limited influence interactions (e.g., Calvin Klein), others show tendency towards mainly receiving (e.g., Jessica Howard) or exerting (e.g., Jones New York) influence. Fashion brands like Ever Pretty have diverse and more balanced influence relations with the rest. Click on a brand's node to highlight its influence relations.
Influence on global fashion trends
Discovered influence by our model of fashion brands on global trends of 20 dresses' fashion styles learned from the AmazonBrand dataset. The width of the connection is relative to the influence weight of that brand in relation to other influencers of the same style.
Related Projects
Fashion Forward: Forecasting Visual Style in Fashion
Ziad Al-Halah, Rainer Stiefelhagen and Kristen Grauman
IEEE International Conference on Computer Vision (ICCV), October 2017.
[paper]
[project]
From Paris to Berlin: Discovering Fashion Style Influences Around the World
Ziad Al-Halah and Kristen Grauman
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2020.
[paper]
[project]