Deep detector

Inspiration

In a world where AI-generated images can be indistinguishable from real ones, people’s memories, identities, and creative work are at risk of being misrepresented. We wanted to create a tool to protect authenticity and help people trust the images they see and share.

What it does

Deep Detector identifies whether an image is real or AI-generated. Users can upload images and instantly get a prediction, giving them confidence in what they’re viewing or sharing.

How we built it

We used EfficientNetB0 as the base model and fine-tuned it on a dataset of real and AI-generated images. The model was trained with data augmentation to improve accuracy and handle various image types. The app is designed to be lightweight and user-friendly for easy integration with web or mobile platforms.

Challenges we ran into

Limited dataset size for training, which made achieving high accuracy difficult. Overfitting due to a small number of images, requiring careful regularization and dropout. Balancing model complexity with performance to ensure the app runs efficiently.

Real-world AI detection is challenging and requires constant adaptation as generative AI evolves. Proper preprocessing and augmentation are critical for small datasets. Building a practical application requires balancing accuracy, speed, and usability.

Built With

• chatgpt
• css
• figma
• html
• javascript
• json
• python
• tensor