|
|
--- |
|
|
title: AI Image Detector |
|
|
emoji: 🔍 |
|
|
colorFrom: green |
|
|
colorTo: red |
|
|
sdk: gradio |
|
|
sdk_version: 6.0.2 |
|
|
app_file: app.py |
|
|
pinned: false |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
# AI Image Detector |
|
|
|
|
|
Detect whether an image is **AI-generated** or a **real photograph** using deep learning. |
|
|
|
|
|
## Model Architecture |
|
|
|
|
|
- **Base Model:** MobileNetV2 (pretrained on ImageNet) |
|
|
- **Transfer Learning:** Custom classification head with: |
|
|
- Global Average Pooling |
|
|
- Batch Normalization |
|
|
- Dense layer (256 units) with L2 regularization |
|
|
- Dropout (0.7) |
|
|
- Sigmoid output for binary classification |
|
|
|
|
|
## Training Data |
|
|
|
|
|
The model was trained on a combined dataset of ~128,000 images: |
|
|
- **CIFAKE Dataset:** 100,000 images (50k real, 50k AI-generated) |
|
|
- **Tiny GenImage Dataset:** 28,000 additional images from various AI generators including: |
|
|
- Midjourney |
|
|
- Stable Diffusion |
|
|
- BigGAN |
|
|
- ADM |
|
|
- GLIDE |
|
|
- VQDM |
|
|
- Wukong |
|
|
|
|
|
## Usage |
|
|
|
|
|
Upload any image to the interface, and the model will predict whether it's: |
|
|
- **Real Image:** A genuine photograph |
|
|
- **AI-Generated:** Created by an AI model |
|
|
|
|
|
## Files |
|
|
|
|
|
- `app.py` - Main Gradio application |
|
|
- `requirements.txt` - Python dependencies |
|
|
- `transfer_model.keras` - Trained model weights |
|
|
|
|
|
## License |
|
|
|
|
|
MIT License |
|
|
|
|
|
## Acknowledgments |
|
|
|
|
|
- CIFAKE Dataset: https://www.kaggle.com/datasets/birdy654/cifake-real-and-ai-generated-synthetic-images |
|
|
- Tiny GenImage Dataset: https://www.kaggle.com/datasets/yangsangtai/tiny-genimage |
