Dermalyze

• 

  Logo

Inspiration:

As teenagers, we understand the common skin conditions that often plague us, such as acne, and how these issues can lead to feelings of insecurity. In today's digital age, an abundance of information on skincare products and routines is at our fingertips. However, this wealth of information can be overwhelming and confusing. That's precisely why we conceived Dermalyze. Our vision is to simplify this process and empower individuals to take control of their skin health and boost their self-confidence. Through a straightforward skin analysis, Dermalyze generates personalized recommendations, offering a tailored approach to improving skin health and radiating confidence. We're here to help navigate the world of skincare with ease and clarity.

What it does:

Our project is built on deep learning, featuring a convolutional neural network that's been trained with a vast dataset of 19,000+ skin images. This neural network can accurately identify 23 different skin conditions. When a user uploads a skin image to our web application, it analyzes the image with over 90% accuracy. If no skin condition is detected, it provides a "normal" result. In addition to this, our application offers tailored product recommendations based on the image analysis. These recommendations are meant to make it easier for users to improve their skin health, all with the aim of simplifying their skincare journey.

How we built it:

Our project was a collaborative effort, with two team members focusing on the deep learning components, and two dedicated to the frontend development. Given the time constraints of the hackathon, creating a novel convolutional neural network (CNN) architecture was impractical. Instead, we extensively researched existing CNN architectures and ultimately selected VGGnet for our project.

The journey began with data preparation, processing, and training, all executed using powerful deep learning frameworks. We fine-tuned our model to ensure it could accurately classify the 23 skin conditions, drawing from our vast dataset.

To offer comprehensive skincare recommendations, we embarked on an extensive research endeavor, carefully curating insights from highly credible sources. However, we later realized that there was an opportunity to streamline this process further by creating an automated product recommendation system, which could have saved us valuable time and effort.

On the frontend, we utilized HTML and JavaScript to craft an engaging and user-friendly website. We seamlessly integrated the frontend with the backend, allowing for image uploads and storage through Flask, resulting in a well-rounded and fully functional application.

Challenges we ran into:

We encountered a couple of notable challenges in the course of our project. During the model training phase, we found that each epoch iteration took significantly more time than we had initially estimated. What we thought would be a quick process turned into several hours for just 10 iterations. Additionally, when implementing the image upload feature for users, we faced difficulties in seamlessly integrating this functionality and feeding the user-uploaded images into our model.

Accomplishments that we're proud of:

Going into this project, none of us had ever worked on a coding project of this size and scope. We all had basic coding experience, but being able to finish this project was extremely rewarding. We’re also really proud of being able to train the AI and utilizing it to create a full scale web application.

What we learned:

Our project journey was a big learning experience for all of us. We gained a solid understanding of CNNs and deep learning frameworks, which turned out to be a strong area of growth. Additionally, we improved our skills in integrating both backend and frontend components, a crucial aspect of our project's development. We also implemented APIs, using them to provide users with the valuable information they needed, such as finding local dermatologists. One of the most significant takeaways was our ability to work well as a team under the constraints and pressures of the project.

What's next for Dermalyze:

Honestly, we want to continue building on our image classification. Our primary goal is to enhance precision, aiming to provide users with even more detailed and precise feedback. For instance, we envision the ability to discern and provide insights on specific types of acne an individual might have. With various datasets available for classifying different acne types, we are confident we can accomplish this. Additionally, we want to create a product recommendation engine that will automate the recommendation process.

Built With

• css
• html
• javascript
• python