Fashion Forward: Forecasting Visual Style in Fashion

Ziad Al-Halah   Rainer Stiefelhagen   Kristen Grauman

[Paper]   [Examples]   [Downloads]

---

Goal

Forecast future fashion trends based on visual data.

---

Contributions

• We introduce the first approach to forecast the future popularity of fashion styles
• Automatic style discovery from large scale data
• Style description via signature attributes
• Analysis of fashion styles lifecycle

---

Paper

---

Downloads

Data sets

• All three data sets used in our work can be downloaded from here

Annotations

• The ids of the items in the three data sets: [Dresses] [Tops&Tees] [Shirts]

Models

• The fashion attribute network: [AttributeNet]

Features

• Fashion elements (semantic attributes): [Dresses] [Tops&Tees] [Shirts]
• Text-based features: [Dresses] [Tops&Tees] [Shirts]
• Tags-based features: [Dresses] [Tops&Tees] [Shirts]

---

Example Analysis

Style discovery Example of some of the discovered visual styles on Dresses, Tops&Tees and Shirts datasets. Our model captures the fine-grained differences among the styles within each genre and provides a semantic description of the style signature based on visual attributes.

Style forecast The forecasted popularity estimated by our model for 4 styles from the Tops&Tees dataset. Our model successfully predicts the popularity of styles in the future and performs well even with challenging trajectories that experience a sudden change in direction like in (c) and (d).

Style life cycle Our approach offers the unique opportunity to examine the life cycle of visual styles in fashion. Some interesting temporal dynamics of the styles discovered by our model can be grouped into: (a) out of fashion; (b) classic; (c) in fashion or (d) trending; (e) unpopular; and (f) re-emerging styles

More... you can find more results and interesting findings in our paper.

---

Related Projects