First step
First, the App will use your camera to analyze the item you are
looking at. Thanks to ml5.js and classification, the system will
be able to recognize it. It will find its sources in a dataset
created by an image classification updated every day thanks to all
the users' pictures.
We need to train our model to recognize different clothes
characteristics, then our model will be able to identify an item
with a certain degree of precision. You can watch this video to
see how to code the web came image classification :
You need to code your classifier like that.
Then you need to use Mobile net, a pre-trained model, and train it
with our own datasets of clothes. We need to teach the model what
is a shirt, what is a dress, and so on. You can visit this webpage
to learn how to train your MobileNet model on the TensorFlow
webpage.
Second step
Then, the App could give options such as shapes recognition to
know if the outfit you like would suit you. In that case, the App
will use some other technology: the same kind of technology as we
have in our iPhones, for example, with the FaceId. As FaceId uses
nodes to recognize your face to unlock your iPhone, our App will
use nodes to analyze your body. You can watch this video to see
how to code body measurements using TensorFlow:
You can also watch this video to easily understand how the
measurement technology would work.
You can also see what would be the result by going on this App which uses the same technology: Find-size
To sum up
This App will help you in your everyday life as a fashion lover.
It could also be a way for brands to be seen all around the world
with no efforts, so it could be interesting to work with brands to
access their datasets in exchange for visibility on the App. It
could be a win-win.
This project aimed to gather fashion marketing and artificial
intelligence. It is a way to renew how to consume fashion and
maybe to reduce the impact on the environment by allowing people
to only buy what they really like and what suits them.