# RTX 4060 Optimized Deepfake Detection This project implements a high-performance deepfake detection system optimized for RTX 4060 GPU with comprehensive evaluation metrics. ## Features - **GPU Optimized**: Specifically tuned for RTX 4060 with 100% GPU utilization - **High Accuracy**: Target >90% accuracy with enhanced model architecture - **Comprehensive Metrics**: - Confusion Matrix - AUC/ROC Curve - F1 Score - Precision & Recall - Cross-entropy (Entropy) - Classification Report ## Quick Start 1. **Setup Environment**: ```bash python setup_gpu.py ``` 2. **Download Dataset** (if not already done): ```bash python datasetdownload.py ``` 3. **Run Training**: ```bash python main.py ``` ## Model Architecture - **Dual-Branch EfficientNet-B0**: RGB + DWT (Discrete Wavelet Transform) features - **Attention Mechanism**: Enhanced feature fusion - **Advanced Augmentation**: Optimized for deepfake detection - **Batch Size**: 16 (optimized for RTX 4060 8GB VRAM) ## Key Optimizations for RTX 4060 - **Memory Management**: Optimized batch size and data loading - **cuDNN Optimization**: Enabled for maximum performance - **Mixed Precision**: Automatic optimization - **Persistent Workers**: Faster data loading - **Pin Memory**: Accelerated GPU transfers ## Expected Results - **Target Accuracy**: >90% - **Training Time**: ~15 epochs for convergence - **GPU Utilization**: Near 100% - **Memory Usage**: ~6-7GB VRAM ## Output Files - `best_deepfake_model.pth`: Best trained model - `final_metrics.png`: Comprehensive metrics visualization - `training_history.png`: Training progress plots - `metrics_epoch_X.png`: Epoch-specific metrics (for best epochs) ## Troubleshooting If accuracy is below 90%: 1. Increase training epochs 2. Adjust learning rate 3. Add more data augmentation 4. Check dataset balance ## Hardware Requirements - **GPU**: RTX 4060 (8GB VRAM) or better - **RAM**: 16GB+ recommended - **Storage**: 10GB+ free space - **CUDA**: 11.8+ compatible# deeplearning