RizzyFits | Devpost

Inspiration

While talking about things that we wished we had, one of us mentioned that he wished he had an app that told him what to wear outside, as he always forgot to check the weather. Inspired by this remark and a desire to become more fashionable, we decided to create (NAME?) in order to satisfy the needs of people choosing what to wear or wanting to try out new styles of clothing.

What it does

RizzyFits is a webapp that takes into consideration the current weather and popular fashion styles to choose an outfit suitable for the temperature and that it doesn't look gaudy or awkward. We are targeting males and male clothing. Users can submit their clothes to our app, and it will store it inside an inventory. We then take into consideration all possible outfit choices from the inventory, and choose the best outfits to show the user.

How we built it

We built the front-end using vue.js For the backend, we used sql to make a database of a user's inventory, hard coded some basic fashion and weather rules, and used machine learning models to analyze which types of clothes fit well together. We also made our own training set of good and bad clothing combinations through scraping multiple online sources. Finally, we got some feedback from outside sources which influenced how we designed our product.

Challenges we ran into

Initially, we were planning to have the user take a picture of their clothes, and use image segmentation technology to automatically find the clothes in the image and detect the possible attributes. However, we weren't able to get a compatible API with our system. This is because many old APIs ran in Python 2 or used esoteric/outdated libraries, and many others were unreliable since they were trained on a very specific dataset/task. We had to scrap this idea and instead allow the users to input attributes themselves. We also had a lot of difficulty finding suitable datasets, especially ones that combined fashionable outfits along with non-fashionable ones. Since we were targeting male clothing, we had to remove the majority of the images in such datasets, since most models were female.

Accomplishments that we're proud of

We're really proud of developing our own dataset: we used an innovative strategy of randomizing different clothing combinations, which would almost certainly result in bad outfits, to give bad outfits. Since many online datasets only gave good clothing combinations, by combining these two, we were able to get a full scope on which outfits were good and which were bad. We are also proud of developing and training our own model on this set.

What we learned

We learned how to adequately plan out our project: coming in, we were quite disorganized, but as we continued, we began drawing out our frameworks, possible strategies/pivot points, and necessary tasks. We also became better at dividing up work and ensuring everyone had a task to be occupied with, throughout the entire project.

As for technical skills, we became better as using SQL, adapting, implementing, and using APIs, reading research papers, searching things up and finding resources, and connecting front end with backend.

What's next for RizzyFits

We hope to eventually get the image segmentation idea to work, and allow users to upload images of their clothes or wardrobe and automatically create an inventory of it. We also will keep iterating on our models, making them more accurate and rizzy. Finally, we plan on getting user feedback and expand to non-male clothing options.