This project aims to classify whether an image is real or AI-generated, using two state-of-the-art deep learning models: EfficientNet-B3 and Swin Transformer. It's wrapped in a professional and user-friendly Streamlit app that includes image upload, prediction, and Grad-CAM-based visual explanation.
- Classifies images as Real or Fake
- Built using EfficientNet-B3 and Swin Transformer
- Real-time predictions via a Streamlit web app
- Visual explainability using Grad-CAM
- Analysis of model weaknesses on stylized art
- Pretrained on ImageNet
- Custom classifier head
- Selective fine-tuning of top convolution blocks
- Grad-CAM integrated for interpretability
- Hierarchical Vision Transformer (via
timm) - Transformer layers frozen initially
- Classification head retrained
- Real Images: Natural photos from open datasets
- Fake Images: AI-generated from models like Midjourney, DALL·E, etc.
- Transforms:
- Resize to 224×224
- Normalize using ImageNet mean & std
git clone https://github.com/xManedge/SpotTheBot.git
cd SpotTheBot```
```streamlit run app.py```