Source Camera Model Identification via Federated Learning using Laplacian-based Patches

This repository contains implementation of various FedFFTNet method for Source Camera Model Identification in Federated Learning Setup. The process involves two major phases:

1. Patching of Images
2. Model Training

📁 Folder Structure

Ensure your dataset is pre-split into train, test folders in the following structure:

data/
├── train/
├── test/

🧩 Step 1: Image Patching

Laplacian based patching including distributing between global server and clients with a separated test dataset has been implemented.

➤ Notebook: Fed_DataSplit_Laplacian.ipynb

Hyperparameters

Variable	Type	Default	Description
datapath	str	./data	Source data directory
number_clients	int	5	Number of clients
initial_ratio	int	0.06	Globally available data ratio
dist_type	str	equal	Data distribution : equal, random, skew

---

🖼️ Patching Overview

Figure 1: Laplacian-based image patching and distribution across clients.

---

🧠 Step 2: Model Training

Once patching is complete, the resulting folder (e.g., patches/) should contain initial (if any) , clients, and test directories.

➤ Notebook: FedLearn_FedFFTNet.ipynb

Configuration Parameters

Variable	Type	Default	Description
root	str	./patches	Directory containing patched data
global_available	bool	True	Indicates if global data is available
batchsize	int	128	Batch size for training
ROUNDS	int	100	Number of global communication rounds
CLIENT_EPOCHS	int	1	Number of local training epochs per client per round
GLOBAL_EPOCHS	int	30	Number of global training epochs
learning_rate	float	0.001	Learning rate
device	str	cuda:0	Device used for training (cuda:i or cpu)

---

🧩 Model Architecture

Figure 2: Overview of the proposed FedFFTNet architecture used for federated learning.

---

✅ Notes

This code will be used for academic and research purpose only. Do cite the paper (Accepted on IEEE Transactions on Artificial Intelligence) if using the method:

@article{chakraborty2025source,
  title={Source Camera Model Identification via Federated Learning using Laplacian-based Patches},
  author={Chakraborty, Rejoy and Goyal, Puneet},
  journal={IEEE Transactions on Artificial Intelligence},
  year={2025},
  publisher={IEEE}
}
---