Federated Deepfake Detection

Introduction

Deepfake detection is becoming increasingly critical due to the rapid growth of AI-generated media. Traditional deep learning methods often rely on centralized data collection, raising concerns about data privacy and scalability.

This project proposes a federated learning framework based on RepLKNet, enabling collaborative Deepfake detection without sharing raw data. By leveraging large-kernel convolution and client-side training, the system aims to preserve privacy while maintaining high accuracy.

---

Dataset

• FaceForensics++
• 140k Real and Fake Faces

---

Directory Structure

Deepfake-Detection
├── data                        # Dataset storage
├── environment.yml             # Conda environment config
├── eval_central.py             # Centralized evaluation script
├── federation                  # Federated learning methods
├── lib                         # RepLKNet and dependencies
├── LICENSE
├── model                       # Training models
├── README.md
├── requirements.txt
├── scripts                     # Training scripts
├── software_engineering        # UI
├── tools                       # tools
├── train_central.py            # Centralized training script
├── train_fedavg.py             # FedAvg
├── train_fedDyn.py             # FedDyn
├── train_fednorm.py            # FedNorm
├── train_fedprox.py            # FedProx
└── visualize                   # ERF computation and visualization

Getting Started

git clone --recurse-submodules https://github.com/l28210/Deepfake-Detection.git
conda env create -f environment.yml

environment : Ubuntu 20.04 + CUDA 11.8 + nvcc 11.8 + cudnn 9.1.0 + python 3.9.0 + pytorch 2.6.0 + gcc 9.4.0 + NVIDIA driver 535.183.01

More details can be found in requirement and environment

• Efficient Large-Kernel Convolution with PyTorch

To use efficient large-kernel convolution with PyTorch, , compile the custom extension:

# compile
cd lib/RepLKNet-pytorch/
unzip  cutlass.zip 
cd cutlass/examples/19_large_depthwise_conv2d_torch_extension
chmod +x setup.py
./setup.py install --user

• central trainnig

# trainning and evaling
./run_central.sh