Learning about some environmental impact of the retail industry led us to wonder about what companies have aimed for in terms of sustainability goals. The textile industry is notorious for its carbon and water footprints with statistics widely available. How does a company promote sustainability? Do people know and support about these movements?
With many movements by certain retail companies to have more sustainable clothes and supply-chain processes, we wanted people to know and support these sustainability movements, all through an interactive and fun UI :)
What it does
We built an application to help users select suitable outfit pairings that meet environmental standards. The user is prompted to upload a picture of a piece of clothing they currently own. Based on this data, we generate potential outfit pairings from a database of environmentally friendly retailers. Users are shown prices, means of purchase, reasons the company is sustainable, as well as an environmental rating.
How we built it
Backend: Google Vision API, MySQL, AWS, Python with Heroku and Flask deployment
Using the Google Vision API, we learn of the features (labels, company, type of clothes and colour) from pictures of clothes. With these features, we use Python to interact with our MySQL database of clothes to both select a recommended outfit and additional recommended clothes for other potential outfit combinations.
To generate more accurate label results, we additionally perform a Keras (with Tensorflow backend) image segmentation to crop out the background, allowing the Google Vision API to extract more accurate features.
We built the front-end with React, using Firebase to handle user authentications and act as a content delivery network.
Challenges we ran into
The most challenging part of the project was learning to use the Google Vision API, and deploying the API on Heroku with all its dependencies.
Accomplishments that we're proud of
Intuitive and clean UI for users that allows ease of mix and matching while raising awareness of sustainability within the retail industry, and of course, the integration and deployment of our technology stack.
What we learned
After viewing some misfit outfit recommendations, such as a jacket with shorts, had we added a "seasonal" label, and furthermore a "dress code" label (by perhaps, integrating transfer learning to label the images), we could have given better outfit recommendations. This made us realize importance of brainstorming and planning.
What's next for Fabrical
Deploy more sophisticated clothes matching algorithms, saving the user's outfits into a closet, in addition to recording the user's age, and their preferences as they like / dislike new outfit combinations; incorporate larger database, more metrics, and integrate the machine learning matching / cropping techniques.