Source Camera Model Identification

This repository contains implementations of various SCMI (Source Camera Model Identification) methods. The process involves two major phases:

1. Patching of Images
2. Model Training

📁 Folder Structure

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

data/
├── train/
├── test/
└── val/

---

🔧 Available SCMI Methods

The following SCMI methods are currently supported:

1. Liu et al. (default)
2. Bennabhaktula et al.
3. Rana et al.
4. Syachanran et al.
5. Huan et al.
6. Rafi et al.

---

🧩 Step 1: Image Patching

All patching strategies are implemented in the patch/ folder, following the parameters defined in their respective papers. You can customize these by passing arguments while running the script.

➤ Script: patching.py

Arguments:

Argument	Type	Default	Description
--src	str	./data	Source data directory
--dest	str	./patches	Destination to save patches
--method	str	liu	Patching method (liu, bennabhaktula, rana, sychanran, huan, rafi)

▶️ Example

To apply patching using the method by Rana et al. and save patches to a folder named patchesRana:

python patching.py --method=rana --src=DATA --dest=patchesRana

---

🧠 Step 2: Model Training

Once patching is complete, the resulting folder (e.g., patches/) should contain train, test, and val directories.

➤ Script: main.py

Arguments:

Argument	Type	Default	Description
--src	str	./patches	Directory containing patched data
--dest	str	./Result	Directory to save prediction results (CSV)
--onlyTest	bool	False	If True, only testing will be performed
--method	str	liu	SCMI method to use (same options as above)
--epochs	int	1	Number of training epochs
--batchsize	int	64	Batch size
--lr	float	0.01	Learning rate
--momentum	float	0.9	SGD momentum
--weightDecay	float	0.00075	Weight decay (L2 regularization)
--modelname	str	bestModel.pth	Name of the saved model file
--log	str	log.csv	File to log training statistics
--numWorkers	int	4	Number of workers for data loading
--device	str	cuda:0	Device to run training (cuda:i or cpu)

▶️ Example

To train using the Rana et al. method for 10 epochs:

python main.py --method=rana --src=patchesRana --epochs=10 --device=cuda:0

---

✅ Notes

• Make sure your dataset is correctly split and organized before running any script.
• You can modify hyperparameters as needed by passing them via command line.
• Logs and results will be saved in the specified --log and --dest directories respectively.

---