In today’s world, visual biometric data is becoming increasingly important, as they are used for various purposes such as generating statistics and training models. However, a major challenge arises because many of these datasets contain identifiable information or personal attributes of individuals. Consequently, regulations such as the GDPR have been established to protect user data, though they also restrict their use. In this work, we propose a body anonymization pipeline that leverages segmentation masks and generative models to replace human bodies in images, thereby concealing individual identities. We show that our pipeline more effectively obscures identities than existing methods while maintaining high image quality.
|
