🔍 Deepfake Detection Using Graph Convolutional Networks (GCNs)

In this project, we tackle the growing threat of deepfakes—AI-generated fake videos that can spread misinformation and violate personal privacy—by moving beyond traditional CNN-based detection methods. Instead, we leverage Graph Convolutional Networks (GCNs) to capture the spatial and geometric relationships between facial landmarks, providing a more nuanced understanding of facial structures.

🚀 What We Did:

• ✅ Built a deepfake detection pipeline that transforms video frames into facial landmark graphs.
• 🧠 Applied GCNs to model facial dependencies, outperforming CNNs in detecting subtle manipulations.
• 📊 Achieved over 90% accuracy, with high precision (92%), recall (89%), and F1-score (90.5%) on the Deepfake Detection Challenge dataset.
• 👁️‍🗨️ Visualized facial graphs, highlighting key regions (eyes, mouth) where manipulation often occurs.

🛠️ Skills & Technologies:

• Graph Convolutional Networks (GCNs)
• Facial landmark detection
• Deep learning and computer vision
• Video preprocessing & frame extraction
• Model evaluation (Accuracy, Precision, Recall, F1-Score)
• Python, PyTorch (or TensorFlow, depending on your stack)

This work demonstrates the potential of graph-based deep learning in media forensics and opens the door to more robust, interpretable deepfake detection systems.

📁 Dataset: Deepfake Detection Challenge (DFDC)

The DFDC dataset is a large-scale benchmark developed by Facebook AI in collaboration with industry partners to advance deepfake detection research. It comprises over 100,000 video clips featuring both real and AI-manipulated content, created using various deepfake generation techniques. The dataset includes:

• Diverse Subjects: Videos of 3,426 paid actors, ensuring a wide range of facial features, expressions, and backgrounds.
• Varied Manipulations: Deepfakes generated using multiple face-swapping methods, including GAN-based and non-learned techniques, to simulate real-world scenarios.
• Balanced Dataset: A mix of authentic and manipulated videos to train models effectively.
• Ethical Considerations: All participants provided consent for their likenesses to be used and altered in the dataset.

This dataset serves as a comprehensive resource for training and evaluating models aimed at detecting deepfake videos. For more details and access to the dataset, visit the DFDC Kaggle page.

🧪 Methodology Overview

Our deepfake detection approach consists of the following key steps:

1. Frame Extraction: Extract individual frames from video samples.
2. Facial Landmark Detection: Use a landmark detector to identify key facial points (e.g., eyes, nose, mouth).
3. Graph Construction: Represent each face as a graph, where nodes are landmarks and edges denote spatial relationships.
4. GCN Processing: Pass the facial graphs through a multi-layer Graph Convolutional Network to learn spatial and geometric features.
5. Classification: Use the extracted features to classify each frame as real or fake.

This graph-based method enables the model to understand complex facial structures and detect subtle manipulations typical in deepfake videos.