vijithsai/cvit-hybrid-deepfake

CViT Hybrid DeepFake Detector

Hybrid CNN + Vision Transformer model for detecting real vs fake images.

Architecture

• —EfficientNet-B0 (CNN)
• —Patch-based Transformer Encoder
• —Feature fusion head

Classes

• —Fake
• —Real

Input

• —RGB image
• —224 × 224

Training

• —Optimizer: AdamW
• —LR: 1e-5
• —Epochs: 7
• —Dataset: DeepFake image dataset

Output

Binary classification (Fake / Real)