DS 3000 Final Project

Setup

1. Create and activate virtual environment:

   python -m venv .venv
   source .venv/bin/activate

2. Install dependencies:

   pip install -r requirements.txt

3. Download the dataset:

   bash download_script.sh

4. Run the project:

   python main.py

Applying Image Distortions

We provide a specialized script (scripts/apply_distortions.py) to simulate real-world transmission and capture artifacts on the deepfake dataset.

What it does by default:

• Type 1 (Compression): Simulates WhatsApp/Instagram image degradation by re-encoding images to a lower quality JPEG in-memory.
• Type 2 (Moiré Effect): Simulates a digital camera taking a picture of a screen by dynamically generating and blending a subtle, randomized Moiré interference pattern over the image.
• Type 3 (Both): Computes the Moiré effect first (capture simulation), followed by the compression (transmission simulation).
• Multithreading: The script automatically utilizes all available CPU cores to process images rapidly.

Usage

1. Testing on a Subset (Recommended First Step) Use the -n flag to limit the number of processed images (e.g., -n 3 for 3 images). When testing, the script outputs to dataset/tmp/ and creates two files side-by-side (_original and _distorted) so you can visually compare the effect:

python scripts/apply_distortions.py 'dataset/raw/Data Set 2' 2 -n 3

2. Processing the Full Dataset Run without the -n flag to process the entire directory. The script will save the results to dataset/cleaned/, preserving the exact folder structure and original file names so your data loaders continue to work seamlessly:

python scripts/apply_distortions.py 'dataset/raw/Data Set 2' 2