The main goal of this challenge is to develop a model that is able to generate outfits for each individual product. This is a particularly challenging task, as the patterns that make an outfit acceptable are really complex. Most of the properties that make products suitable to belong to the same outfit typically do not exist as product metadata: can a floral short dress be paired with white sandals and a yellow tote bag? The answer is: I need to see the products first.
In fashion, it is the intrinsic visual details and patterns of each product in an outfit that determines the outfit's quality. Being able to properly extract such features from different products and model which of them should be matched together is the demanding challenge we are proposing here. We will be focusing the challenge on the widely known task of Fill in the Blank, that consists of predicting what the missing item of a real outfit is, out of a list of candidate products. This task represents a good approximation of the original problem, as a model that is capable of understanding what should be the style characteristics of the missing product of an outfit can be easily adapted to the generation of full outfits.
For this challenge, the participants will have access to real outfits produced at FARFETCH by stylists and fashion experts. FARFETCH is the leading platform for online luxury fashion shopping, with the biggest catalogue of luxury items in the World with more than 3 million products and more than 10 thousand brands and high-end designers.
Data
The dataset provided for the challenge consists of a list of outfits described by the products that compose it, and the FARFETCH product images and metadata, and outfits composition. This dataset is generated by FARFETCH and provided to the participants in a ready to use manner. It is composed by three data sources:
| Rank | Participant team | FITB |
|---|---|---|
| 1 | Depop | 0.799 |
| 2 | Outfit>Overfit | 0.764 |
| 3 | SefaMerve Research | 0.680 |
| 4 | ColdWheels | 0.660 |
| 5 | lambino | 0.631 |
| 6 | RATC | 0.552 |
| 7 | CUFE | 0.411 |
| 8 | ConsciousAI | 0.175 |
| 9 | jaydo | 0.152 |
| 10 | Luxury A.I. | 0.089 |
| 11 | uva | 0.049 |
| 12 | seasee | 0.015 |
You can find the Data Challenge here.
Training data, evaluation scripts and rules can be found in the challenge page. In addition, a more detailed description of the dataset is also provided in that page.
The first placed team will receive 1250€, the second 500€ and the third 250€, all in Farfetch vouchers to spend freely at Farfetch.com.
Please note that the prizes are attributed to the teams, and team-members should split the voucher value as they see fit.
Paper submissions will be open from the 11th of June. Please use this link to submit a paper presenting your solution.
The deadline for paper submissions is the 27th of June. The papers should be between 2 and 8 pages. All submissions must be formatted according to the latest ACM SIG proceedings template available at http://www.acm.org/publications/proceedings-template (LaTeX users use sample-sigconf.tex as a template).
| May 3 | Registration/stage 1 starts |
| June 10 | Stage 1 closes |
| June 11 | Stage 2 opens/paper submission opens |
| June 17 | Stage 2 closes |
| June 27 | Paper submission closes |
| July 7 | Paper notifications |
| July 10 | Camera ready submissions |
| July 15 | Workshop |
For more information, please contact us at info@ffrecschallenge.com.