AI Generated Image detection using a lightweight, highly accurate deep learning model. Trained on dataset with over 120k images from various sources like DALL-E,Midjourney,SDXL, etc.
🌟
A lightweight, high-accuracy deep learning model for detecting AI-generated vs real images. Trained on 120,000 images from diverse sources (MidJourney/SDXL/DALLE/camera photos) using EfficientNet-B0, achieving:
- 98.5% accuracy with 120,000-image test set
- Exaplainability using gradcam heatmaps
- GPU for fast inference
- Clean training pipeline optimized for Windows + CUDA
🚀 Features:
Detects AI-generated vs real images
Lightweight and efficient EfficientNet-B0 backbone
Trained on 120,000 images (balanced)
CUDA-optimized PyTorch training pipeline
High accuracy & robust performance
Grad-CAM explainability for visualizing model decisions
Confusion matrix + classification report included
📊 Results Accuracy 98.5% Precision (Real) : 0.99 Precision (Fake) : 0.98 F1 (Real) : 0.99 F1 (Fake) : 0.99 Test Images 20,000
Confusion matrix:
GradCAM Examples:
