ShazaamStyle.




Hello, again. For this third "Designing with Web" project, I have chosen to work on a new application of Artificial Intelligence.

Curious ? Let's get started !


I am sure you've all experienced this annoying situation. You're in the subway, thinking about your day, until you fall in love with the dress that the girl next to you is wearing. This is the dress you didn't know you needed until now. And you have no idea where to buy it.

SO, WHAT DO YOU DO? Start running after this girl in the corridors to figure out the name of the brand? That would not be very appropriate... Well, as you may have guessed, this situation happened to me yesterday, and I thought It would be amazing if I could just take a picture of this dress and instantly know where it comes from.

That is why I started to think. I am not a coding expert, but there's something I know really well: the fashion world and its innovations. And there's one innovation that cannot be ignored: AI.

THAT'S THE SOLUTION !!

So, my investigation began. I started by reread an article found on Vogue Business about how Facebook is using AI to provide outfits recommendations. That didn't solve my problem, but it gave me a few ideas of what my App could do.



The key features

Find the model you're looking for in one shot
Find useful details
Rate your product
Share your findings






Our software reads outfits images submitted by users. It can identify details including silhouette, texture, shape, colour and pattern.

You have access to all the informations: brand, price, collection, production process...

You have the possibility to rate your experience with AI and give your recommendations.

Finally, share your treasure with your friends via Instagram, Facebook or Pinterest!




So, how does it work?

First of all, we will need to implement a classification of clothes. To find these categories, I've been on Asos, as I know that they have a very complete collection of items. Here are the categories I've found:

  • Activewear
  • Skirts
  • Dresses
  • Hoodies
  • Jeans
  • Jumpers
  • Shorts
  • Coats
  • Tops
  • Trousers
  • Jumpsuits
  • Swimsuits
  • Boots
  • Sandals
  • Trainers
  • Heels

Considering the nature of our AI, our machine learning process will use unsupervised learning. We have to establish and give a large dataset with classifications to the machine. The dataset will contain many clothes for each categories we defined and the AI will learn from that data.

Feed the AI.

Ml5.js is a library which provide access to machine learning algorithms and models in the browser. MobileNets are a class of convolutional neural network which are trained to recognize images. The AI will try to compare the image with all of the images which are on its image database. SO, in our case, we need to set up ml5.js and MobileNet to classify an image, and to teach to the AI many different images of what we want it to recognize (that is to say, clothes). To be efficient, we will need to find many pictures so that the AI could learn all the possibilty based on the ABCDE system. Once we have all the image we need, we will have to insert those images. Here's a video showing how to get our own image database:


As an example, here's what MobileNet will show us once our own database is created:


As you can see, for this sweatshirt the AI is 96,8% sure that it comes from Zara, at a price of 15,99$.
For our project, we really need to add the live video. Thanks to that, the AI will be able to recognize an object or something else when we will be filming. There is a code that we can use to add the live video :

When we add the live video, we can show to the camera a piece of clothing and the AI will recognize it and tell you what it is.

Limits

As you may have noticed, this project is really ambitious. Fashion is changing nonstop, that is why our AI has to learn fast and our database has to be huge. Brand's logos should also be included in the database in the future, in order to improve recognition.




Beyond the project: the opportunities

In the future, a good thing would be to create partnerships with famous (or unfamous) fashion brands so that customers can buy directly the products they want on the brand's website. Going further, the AI could also provide recommendations of similar things or items that can match with the item searched. In this way, you will never be out of style !