Detects AI-generated vs real images using a multi-domain deep learning approach — combining RGB, gradient (Sobel), and FFT features with a ResNet18 backbone. Achieves 95.05% accuracy on the CIFAKE dataset. Includes Grad-CAM explainability visualizations.
computer-vision deep-learning pytorch image-classification fft transfer-learning binary-classification gradcam fake-image-detection explainability resnet18 ai-detection cifake multi-domain-features
-
Updated
Mar 6, 2026 - Python