-
Our application in action!
Inspiration
AI has started to become incredibly advanced, being able to create even flawless photos from a simple text prompt. However can AI match pace with human creativity? Art is considered the purest form of human creativity and is designed to capture a human's attention. However, can AI generate art that outdoes these humans who have spent their lives training their art skills?
What it does
The web applications functions as an experiment collecting data to answer our hypothesis: Can AI art compete with human-made art? There are 7 trials, and in each trial, a pair of one AI art and one human-made art is shown on screen for 10 seconds before moving on to the next pair. To control for outside factors, the website uses the same color background for both sides of the screen, and each pair has a similar style and subject. The web application opens a video tracker to take the dimensions of your face and eyes. Image cropping, filtering, and analysis are used to calculate which direction you are looking. The direction of the gaze for each frame is recorded. Whichever image received the most amount of frames where the user is looking in its direction "wins" the trial. Whichever group, the AI or the humans, wins the majority of the trials is then declared the winner at the end of the experiment.
How we built it
To create this project, we divided the work among the three of us. One person would take on the task of web scraping and data collection, another would deal with web design, and the third would work on the program to track eye movements. Prior to any program, we spent a good deal of time installing any necessary dependencies and familiarizing ourselves with the libraries we would be using. As well, we also got the basics of GitHub and proper commit policies among us worked out. The initial plan was to implement the completed eye tracking script using opencv into a web application, and although we were successful in creating an eye tracker that worked reasonably well, due to our inexperience with web design tools such as flask and react, we realized that implementing this into a web application would not fit within the time frame. Thus, we settled to create a client-based application using the simple GUI library known as tkinter paired with threading to run both the eye tracker and the GUI simultaneously. Once we had a working application, we would set about tweaking it to provide the best user experience, making any necessary compromises along the way.
Challenges we ran into
We ran into several setbacks from the start of the project. Of course, we had the standard problem of having insufficient knowledge, which required subsequent research. But there were also specific challenges that stood out. First, the python module that we planned to use to track eye movement wasn't completed and the alternative provided very raw data. We also faced trouble getting the images that we planned to display to the user. Although we had fully functional code that allowed both autonomous ai art generation and web scraping, we repeatedly had to find new ways to complete these tasks due to problems such as faulty APIs and needing to pay for authentication keys. One of the biggest challenges, however, was connecting the python program to the website. We were attempting to use Python Flask, however, it was difficult to keep this process updated, and called multiple times since it couldn't run dynamically. In the end, we had to resort to attempting to write a JSReact program along with a supporting Python GUI.
Accomplishments that we're proud of
We are proud that we persisted and finished this ambitious project. We are also all really proud of ourselves for learning so much new technology in one day to make this project possible, such as computer vision, react, and web development.
What we learned
Andrew Zachary Liu - "I've learned so much about web development, such as Flask, dynamic vs static changes, and React" Andrew H. Liu - "I learned much more about how to use libraries such as opencv and tkinter in python. As well, I have discovered the importance of threaded processes in python." Dhruv Saxena- I learned a lot about web scraping, APIs, and web development through React and Flask frameworks.
What's next for ART - AI Retina Tracker
Since we were low on time, we couldn't finish everything we set for. We plan on adding more trials and more features to control for variables, such as picking images with similar dimensions and randomizing the side AI and human art is displayed on the website. If possible, we want to expand this project and collect data on a greater population so that the data can possibly be used by others to answer research questions.
Log in or sign up for Devpost to join the conversation.