Use Cases and Challenges of Ethical Offline Anonymization with YOLOv8: PixelMe on CPU-Only Systems

This paper presents PixelMe, a privacy-preserving, fully offline system for high-accuracy facial and object anonymization based on the YOLOv8 deep learning architecture. PixelMe operates entirely on local hardware with no cloud connectivity, integrates FFmpeg and a CPU-optimized detection pipeline, and provides a graphical interface accessible to non-experts. The system was evaluated on low-resolution and high-resolution video datasets under CPU-only conditions, demonstrating reliable multi-class detection, effective anonymization, and near–real-time performance. PixelMe addresses gaps in existing tools by combining usability, data minimization, and privacy-by-design principles within a practical anonymization workflow. The platform supports GDPR-compliant documentation and enables secure handling of sensitive visual data in inspection, journalism, and industrial contexts.