OSME | Devpost

User interface for OSME

Inspiration

The inspiration for our project comes from our team's desire to use AI for good. Nowadays, online shopping is so convenient that many people prefer to purchase clothes that way rather than in-person. The shipping of packages makes up around 3% of global CO2 emissions and shipping can make up as much as 10% of greenhouse gas emissions by the year 2030. The environmental impact of buying and returning packages is growing and our project seeks to minimize this impact.

What it does

OSME is a web application that uses AI to predict whether or not an article of clothing will fit the user before they make a purchase. Our application takes the user's body measurements such as torso or arm length and determines if the clothing fits or is too loose/tight.

How we built it

We built our project by generating our own fake data but using realistic measurements rather than ones that are purely random. We created measurements that will and won't fit a person so that our model can train on more diverse data.

Challenges we ran into

One major challenge we ran into was determining how and where to get the data to train on. We initially created synthetic data with a Python script but it was too random to be useful. We then found an online dataset specifically for this kind of problem but it was missing a lot of data we needed. Another challenge we ran into was getting a 100% accuracy for our model which is an unrealistic metric to attain.

Accomplishments that we're proud of

Our team is proud of conceptualizing a project that uses AI to help make other people's lives more convenient while also making a positive environmental impact. We're also proud of learning about the different technologies involved in the process of a machine learning project when our team members had no previous AI experience.

What we learned

We learned that the process of collecting data for a machine learning model is not a trivial task. We learned that there are many different variables to take into account when gathering data such as how accurate it is, whether or not the data is diverse, data normalization, etc.