Machine learning for women’s closets

At Chicisimo, we believe that clothes in our closets will soon be digitized, and this will change everything.

We are building an automated personal stylist that learns your clothing habits. It then helps you decide your daily outfit, and tells you how new clothes match your existing ones. In order to achieve this, we are focused on three jobs: capturing data, cleaning data, and developing the right algorithms to help people.

Deep learning algorithms are great at fashion detection, and at matching fashion items with similar items. But no algorithm can match products to people, because there is no people data. This is a fascinating problem, which we want to solve.

Our enabling infrastructure contains four assets:

1. A consumer app where people are storing their clothes, and seeing outfits uploaded by other people wearing those same clothes. Read more below;

2. A data platform that receives the data provided by people through our mobile app. With this data, the platform produces a dataset of clean, structured and correlated data that technology can interpret, and offer services on top of it. This platform is called the Social Fashion Graph, and it includes a fashion ontology that plays a key role. Read more below;

3. A dataset of correlated descriptors, outfits and people. These descriptors are a list of people’s what-to-wear needs, expressed in different forms. This dataset is exposed via a dataportal, which provides us with transparency and control. Read more below;

4. An IP portfolio protecting our innovations: tagging images with shoppable products; extracting correlations among clothes, in outfits and closets; outfits search. Read more here.

The above infrastructure is solving the single biggest problem people have with their clothes: How to wear them. There are other use cases, but all related to this problem. It affects people who absolutely know how to combine their clothes, and also people who do not; it affects fashion lovers, or people who simply need to decide what to wear.

In the last few years, deep learning has seen impressive progress. As a result, we see how ecommerce companies are slowly converting their clothing catalogues into graphs of correlated clothes and descriptors (descriptors defined by experts and attached by machines).

Having reached this stage is a great foundation for what’s next: including people's tastes and closets into those graphs, enabling true artificial intelligence fashion stylists and AI personal shoppers. We will then see how artificial intelligence will impact the future of fashion.

The Social Fashion Graph

The Social Fashion Graph is the name we’ve given to our data platform. It learns about people’s what-to-wear needs, and attaches those needs to outfits and to people.

The backbone of the Social Fashion Graph is our ontology, which plays a critical role at giving structure to the incoming data.

We’ve learnt that the top need we all have is how to wear the specific clothes that we have in our closets, together with everyday special occasions, adapted to our characteristics. Dealing with these needs is the job of the Social Fashion Graph, once data is transfered from the app.

The result of the above is a dataset of clean, structured and correlated data that technology can interpret.

We’ve exposed all this data thru a dataportal, which has provided the team with transparency and control. As a result, it’s easier to understand where we are, and what’s next.

We do think that machine learning and deep learning in fashion are going to make people’s lives better. We are focused on the entire infrastructure and operations to ship this to people, and on having people be part of the infrastructure.

Our consumer app

Chicisimo’s iPhone and Android apps connect the Social Fashion Graph to the reality of people’s needs, and captures clothing data.

Most important of all, the apps help us learn, they bring unique impact to our learning process. Thanks to our app, we receive daily direct feedback from many people, which helps us learn.

We think this is the most interesting aspect of building a consumer product. The fact that, regularly, we access new corpuses of knowledge that we did not have before. This new knowledge helps us improve the tech and product significantly, and it is a great reminder that we are not in the upper part of the learning curve -we are simply moving up.

When we’ve obtained these game-changing learnings, it’s always been by focusing on two aspects: how people relate to the problem, and how people relate to solutions.

Iterating a consumer app is a unique learning experience.

You can read about our learning process in this Medium piece, and see how we deal with retention, onboarding, etc. And in this link, you can read Apple’s description of Chicisimo (“your dedicated personal stylist— always on-call, living in your handbag”)- they feature our app regularly in more than 60 countries. Hey, we had to mention it! And finally, we use this sentence for a test we are running :): clothes app, outfit planner app.

Usamos cookies propias y de otros para ofrecerte una mejor experiencia de usuario. Al utilizar nuestros servicios, entendemos que aceptas el uso que hacemos de las cookies.