The Real-World Faceswap Dataset addresses a critical gap in deepfake detection research by providing the first comprehensive collection of faceswap content that reflects actual real-world usage patterns.
Overview
Traditional deepfake datasets primarily consist of laboratory-generated content that fails to capture the complexity and post-processing techniques commonly used in real-world scenarios. This dataset bridges that gap by collecting authentic faceswap content from popular online platforms.
Dataset Composition
Content Sources
• Popular online faceswap platforms
• Various quality levels and processing techniques
• Diverse demographic representation
• Multiple generation models and techniques
Technical Specifications
Format: MP4 videos and JPEG images
Resolution: Various (240p to 4K)
Duration: 1-30 seconds per video
Compression: Multiple compression levels reflecting real usage
Key Features
Post-Processing Variety
• Super-resolution enhancement
• Noise reduction filters
• Color correction and enhancement
• Various compression artifacts
Annotation Details
• Generation method labels
• Quality assessment scores
• Post-processing technique annotations
• Authenticity ground truth labels
Research Applications
This dataset enables research into:
• Real-world deepfake detection robustness
• Post-processing impact analysis
• Cross-platform generalization studies
• Practical deployment evaluation
Access and Usage
The dataset is available for academic research purposes. Commercial usage requires separate licensing. Please cite our accompanying paper when using this dataset in your research.
Ethical Considerations
All content was collected with appropriate permissions and follows ethical guidelines for synthetic media research. Personal identifiable information has been appropriately handled according to privacy regulations.
