Need Fashion Advice? Ask an Algorithm
A scientific paper from University of Toronto and Spain’s Institut de Robòtica i Informatica Industrial shows how machine vision will be able to tag a fashion faux pas
An open-access paper from the Computer Vision Foundation shows the possibility of using algorithms, machine vision and crowd-sourced data to create a gauge for fashionability. The system can even be used for generating suggestions to improve your fashion sense.
In the paper entitled “Neuroaesthetics in Fashion: Modeling the Perception of Fashionability,” four scholars explored how advanced computer recognition can make sense of an #ootd photo, predict the reaction of a fashion-minded crowd and provide ideas for improvements.
The team from University of Toronto and Spain’s Institut de Robòtica i Informatica Industrial analyzed over 144,169 images in the study. The image set, termed Fashion144k, were collected from Chictopia.com, a website popular amongst the trendy fashionistas.
Using the advanced cloud-based visual recognition capabilities of Rekognition, the team was able to extract key information from the photos. Each time a photo goes through the study, age, race, gender, garments worn, colors, location and website popularity in terms of followers, are extracted. Because of the advanced capabilities of the recognition software, and the photographic style are also taken into consideration. Many of Chictopia images are not just selfies but the background, poses and even time of day play a part in what makes an image fashionable, the research says.
As proxy for fashionability, the team used “user likes” to gauge positive reaction towards the images. Country of origin, user following and time of upload were also taken into account, allowing data to be more reliable in gauging just fashionability.
The paper concludes:
This is an important first step to be able to build more complex and powerful models that will be able to understand fashion, trends, and users a whole in order to improve the experience of users in the modern day society.
We have made both the dataset and code public in hopes that this will inspire other researchers to tackle this challenging task.
Commercial applications are plenty. In a connected world where social and retail places are merging, the process can help brands reach out and give personalized advice to their customers minus the expense.
Fashion design via Shutterstock