Bread Crumbs

Inspiration

In the age of easy AI-generation and image editing it's more common then ever to come across falsified or misleading images. This is why we decided to help combat misinformation by creating Bread Crumbs!

What it does

Bread Crumbs takes an image uploaded by the user and attempts to verify its authenticity. First, it scrapes its EXIF data to look for common features of real and fake images such as GPS data for real images, and keywords like "Photoshop" or "Midjourney" in the falsified images. Next, the model is pushed through an image recognition model trained on top of resnet50 to classify the image as real or fake.

How we built it

We built a simple react frontend for the webpage to allow users to upload their images they want to test. The backend is built with python to allow us to utilize pytorch for our model and the image metadata reading and parsing libraries. The model itself is trained on top of resnet50 with a dataset of three image classes real, AI, and spliced.

Challenges we ran into

The main challenge we ran into was allowing the model to perform well on images found outside of the datasets we trained it on. We found that our dataset was too easy for the model and it was classifying based off of dataset fingerprints rather than actual features of the classes.

Accomplishments that we're proud of

We're proud of our websites look and the fact that we can somewhat confidently classify images.

What we learned

We learned a lot about what it takes to make a successful model for image classification tasks as well as about image metadata and how that can be used to reveal information about an images origins.