While my friends and I were walking around, looking at different sponsors, we found an abandoned booth with stickers at the desk. We wondered about what company it belonged to and there's where the idea spawned—an app that could recognize and categorize different logos, recognizing any one would we could feed. While at the surface this seemed simple, a deeper look revealed the neural networks and training data sets that had to be researched.

What it does

Our web application runs a server on a local computer where it takes in any image that the user enters, and uses a neural network that is pre-taught on training images gathered from the internet in order to classify the entered image. It then produces the percentage of similarity to each of the learned categories, ranking them in order from most similar to least.

How we built it

I used TensorFlow in order to utilize Google's Deep Learning algorithm and employ it in my own application. Andy, one of our team members, then coded the back-end server side of the web application so that users could access the localhost from their phones and upload images. Their image would in turn run through my algorithm and output a results text file, which our third team member, Will, displayed on the front end the results.

Challenges we ran into

Our first challenge was creating an idea because this was only my second hackathon and actually two of the members' first hackathon. This led to a rough time at the beginning as we wasted many hours simply trying to come up with a project idea. The next challenge, once we had an idea, was actually implementing Tensorflow and getting it to cooperate with my computer. Finally, and these are ongoing challenges, HTML does not allow call of local files, which prevents our application from displaying the user's image in our output page, and our configuration of the DNS to a domain we bought refuses to work.

Accomplishments that we're proud of

We got the AI component of the project to work, which is a massive accomplishment in a few hours. We managed to integrate all of our separate codes with little to no compilations.

What we learned

I personally learnt a lot more about neural networks and Tensorflow in particular because I solely focused on that part. Will learnt new web development tools and the basics about creating the front end of a website—the HTML, CSS, as well as some JavaScript. Andy learnt server side tools and how to seamlessly move through pulling user information to running it through the backend to returning it back to the front end.

What's next for neuralogo.

As of right now, neuralogo consists of seven different categories of classification and suboptimal numbers of training images per category. As a result, the classification isn't always accurate. We plan to acquire more training data for each category as well as dramatically increase the number of categories so that we can make this an official logo recognizer that can be popular worldwide. Also we plan to eventually create Android or IOS applications that utilize neuralogo so that people can use our technology.

Built With

Share this project: