HackTheCloset

• 

  Interface

Inspiration

The convenience of e-commerce inspired us to embark on this project, aiming to expand our knowledge and push our boundaries. Our inspiration stemmed from typical ranking systems, with the goal of developing metrics to help users find the perfect piece of clothing quickly and efficiently, thus enhancing the online shopping experience.

Upon hearing about the challenge proposed by Inditex, we knew we couldn't pass it up. It represented not only a growing opportunity but also a fun and rewarding experience.

What it does

Our project features a dynamic screen displaying ten images. When a user clicks on one of the images, the displayed options are adjusted to best match the user's selection, recommending items that we have deemed similar and potentially more appealing.

How we built it

Our journey began with a fundamental question: What defines similarity between clothes, and how can a computer recognize such similarities? This led us to focus on the dataset itself, requiring us to filter it based on features extracted from the URLs and apply preprocessing techniques essential for subsequent steps.

Since the images not only showcased clothes but also people wearing them, we decided to streamline the process by removing images containing humans and backgrounds. Drawing from our familiarity with AI, we utilized feature extraction models to recognize similar patterns.

Our focus extended to color, and we developed the idea further by leveraging built-in modules capable of not only identifying the main color of a piece of clothing but also providing a palette of color information.

Challenges we ran into

The dataset presented the most significant challenge for us. We meticulously analyzed the dataset, corrected faulty rows, and removed broken links.

To better manage the vast amount of data and facilitate model training, we constructed a data frame to systematically organize the products. This categorization proved invaluable in handling the data effectively.

Accomplishments that we're proud of

We are proud of having all pitched in with our ideas and each one of us having put a bit of our style in this project. From troubleshooting the recommender feed to the excitement of training the dataset through manageable batches to prevent any mishaps, we shared many memorable experiences throughout the project. These challenges definitively took a toll on us, but above all, the memory of supporting each other when needed will overshadow them all!

What we learned

We got the hang of handling data with efficiency and safety while having fun along the way! We were constantly on the lookout for solutions that were not only efficient and computationally feasible but also scalable for new targets and new data.

What's next for HackTheCloset

See our service on the actual website! We'd love for clients to test their personalized closet experience.

Moreover, we believe that we can make the user experience even more enjoyable by adding a fun and engaging twist. We're considering gamifying the experience to encourage better engagement and enhance the overall retail experience.

Built With

• flask
• javascript
• python
• react
• tensorflow