Inspiration Currently, society faces many environmental challenges with the fashion industry known as one of the most polluting industries in the world. Fast fashion produces clothing that isn't made to last since they are made with cheap materials that harm the environment (landfill impact, pesticides in growing cotton, and toxic chemicals making their way into water). However, sustainable clothing use materials that are made to last longer and are non-harmful to the environment. By choosing to have sustainable clothing, one can reduce their waste significantly and spend less money when shopping for clothes. We believe that users would benefit greatly from an intuitive app that can be used to track their wardrobe that can let them know how sustainable their wardrobe is and also predict the type of clothing they have using Tangram.

What it does The app allows users to explore their closet by letting the user add clothing that they already have in their closet by selecting the type of clothing (shirt, pants, shorts, jacket, etc.) and then picking the brand of clothing. The user can then take a picture of the clothing item that they wish to upload and can add more items if they wish to. After they are done adding their items, the app then generates their sustainability score and gives them outfit recommendations as well that are more sustainable for their closet, which would take them to the store’s website and purchase it on-app. Users can also get points from buying from a sustainable brand which they can redeem through gift cards from some of their favorite sustainable brands. The model that was built using Tangram recognizes the type of clothing the user has based on what the user takes a picture of.

How we built it For the model: we learned how to work with Tangram, using a CSV to create a .tangram model, and then using that model to test individual data and judge the overall accuracy. Then, we were able to find a large dataset online that had thousands of images of different clothing, all labeled for type (i.e. shirts, shoes, pants). We converted those images into binary strings and created a new CSV with those strings and the corresponding types. That was then used to train a Tangram model and we came out with about 61% accuracy, which is slightly better than a completely random guess would be expected to be, with so many clothing options.

Challenges we ran into

Tangram does not have support yet for image recognition, so we did have to think of a way to pass in the images to test the library in use of that. When training, there were issues in the inclusion of commas in the binary strings (messing up the CSV formatting) and the cleansing of data we performed to remove clothing we deemed irrelevant.

Accomplishments that we're proud of

Although the model is not the most accurate, we are proud of trying to find a way to apply this tool, Tangram, to a new purpose in image recognition. Also, learning how to train and use this machine learning library was a useful skill that multiple people on our team can use in the future.

What we learned

We learned how to use an intuitive machine learning library and a little about how image recognition and data cleansing work.

What's next for Sound of Sustainability

The development of a full front-end using some cross-platform tool such as React Native, and the connection of that with our user interface and machine learning model to create a fully functioning app.

Built With

Share this project:

Updates