The color, size or design are aspects that always generate doubts when buying a garment online, whether it is a pants, a shirt or a dress, making us wonder if at the time of using it this be flattering to our silhouette.
That is why Facebook is working on the development of an artificial intelligence that in the future can assist users throughout their experience in buying clothes online that they make through its set of messaging platforms and tools ( WhatsApp, Messenger, Instagram) in addition to Facebook itself.
This artificial intelligence could provide the user with product recommendations on the fly, as well as learn from their preferences through an analysis of images of their wardrobe in which it would also allow them to try pieces on virtual replicas of themselves.
Through the Conference on Computer Vision and Pattern Recognition (CVPR) 2020 Facebook introduced a series of articles that were accepted and in which it was suggested that the company was in plans to develop the structure of that assistant.
One of these articles describes an algorithm that deciphers and quantifies the incidence of fashion through the use of images from around the world.
In another article an AI model is exposed that would have the ability to generate 3D models of the user from their individual image, while in another article the creation of a system that registers the affinity of clothing based on different forms is proposed. of the body.
Deciphering fashion and its styles
Alluding to what was mentioned by an engineer from the AI research on Facebook, people’s clothing decisions are usually influenced by factors such as comfort, taste and occasion, along with others of less incidence such as changing social norms, celebrities, style icons, politics, art, weather and the prevailing mood in the region.
For all this, it is a real challenge for an artificial intelligence to decipher in quantitative terms the influences that can govern the fashion of one or more individuals.
In this sense, the Facebook researcher proposes to obtain a pattern of influence in fashion by resorting to the large photo galleries. Once the pattern is established, it should be used to determine style trends.
It begins by generating a vocabulary of visual styles based on images of unclassified, geolocated and timestamped people.
Each style is a mix of detected visuals in which, for example, one can record short flowery dresses in bright colors and another capture high-necked shirts.
Then a record of the past trajectory of the popularity of the styles is made in order to help identify the novelty and the temporal origin, a term used to refer to the change that occurs in the fashion of a city before the observed influence manifests itself.
Subsequently, the calculation of the degree of influence between cities is made from a statistical measure, while an AI model deciphers the photographic relationships in order to predict future popular styles in any region.
Rendering of 3D people
Another of Facebook’s articles exposes the creation of an AI-based system that can be used to generate 3D models of dressed people, thus becoming the central pillar of a potential fashion assistant powered by Facebook.
Such a system would be called the Animated Reconstruction of Clothed Humans (ARCH) and would give users the opportunity to appreciate the piece of clothing they wish to buy on a virtual model of themselves in different poses and ranges of motion such as standing, walking, sitting or crouching, as well as in various environments and lighting.
Choosing the right dress
Going into the third Facebook article, we find the proposal of a researcher and a Facebook co-author who propose the creation of a program called Visual Body-aware Embedding (ViBE) which would have the purpose of establishing which pieces of clothing are flattering for the silhouette. of a person depending on the proportions of his body.
There is no doubt that all these proposals represent a good omen towards the materialization of what will be Facebook’s fashion assistant in the future.
However, to facilitate this process, Facebook believes that advances in aspects such as language understanding, personalization and social experiences are necessary so that an optimal and efficient predictive style assistant can be developed.
