Binary Translator

Inspiration

Not many people know that the binary written on the walls of the WFIC is the binary translation of Thomas G. Clemson's will. We wanted to make an app that an everyday person could use to see the reality of the translation of the will.

What it does

The user takes a picture of an 8 bit section of the wall, and a character is displayed as to what that character represents.

How we built it

We used the Watson Visual Recognition service to classify the images taken by the user to be individual characters. Then this response is returned and displayed through the app. The app was written in Java using the Android Studio.

Challenges we ran into

Generating enough images so that we could properly train Watson with a custom classifier. The classifiers we created at times were unreliable and exuded extreme confidence in the wrong answer. Watson's Visual Recognition Service was in Beta at the time of this demo, and the service ultimately broke down, stopped working, and removed all our classifications. Being a proprietary service, we could not debug the root cause of this problem. Because this was the keystone of our entire build, the android app would crash upon sending a request for classification from Watson.

Accomplishments that we're proud of

We successfully created unique training sets through software which ultimately enabled us to create classifiers which could define the individual characters from a string of 8 bits without the need to take over 2000+ pictures of the wall. Our app was able to classify the features of a multitude of objects such as mice and people.

What we learned

We learned how to train custom classifiers in Watson's Visual Recognition service. We also learned java and Android App development for this project.

What's next for Binary Translator

We want to rely on script files to create custom classifiers rather than the Visual Recognition GUI provided by IBM. As we develop this project more for other classes, we hope to increase the number of bytes which can be translated per image, the frequency that these translations appear, and to ultimately port this implementation to an augmented reality headset.