Online Outfitters

Inspiration

An attempt to improve Computer Science students' fashion and clothing wardrobes.

What it does

The website has 3 modes: Upload, where the you can upload an image of a piece of clothing and name it. Once submitted, our neural network will detect what type of clothing and color it is and add it to your clothing library. Library, this is where you can view your already added clothes. Generate, the main feature of the website, allows the user to generate outfits based on their library and a dataset of fashionable colour combinations.

How we built it

Developed and trained a neural network from scratch using PyTorch for detecting the type of clothing the input image was. Then implemented K-means clustering to identify the colour of the piece of clothing inputted. After that we setup the optimal colour combinations with that particular piece of clothing and setup API interactions with LLMs to create 3D models wearing the input piece of clothing and other recommended pieces.

Challenges we ran into

Obtaining clean data for training our neural network. Setting up API interactions with a suitable LLM.

Accomplishments that we're proud of

Building our own neural network from scratch. Full stack development under time pressure. Successful training of the model

What we learned

Training and developing the neural network taught us a lot about the ML pipeline from finding clean data and tuning the model to processing each epoch. How implementing a K-means clustering algorithm can sometimes be the best choice for time complexity. Learning how to creation interactions between API keys.

What's next for Online Outfitters

Increasing the library categories to more detailed types of clothes e.g. tank tops and jeans and expand the colour palette to more than just the primary and secondary colours. Implementing a clothing suggestions feature that recommends you the cheapest clothes missing from your wardrobe with links to purchase. Generating