Our method obscures identity while preserving attributes such as gaze, expressions, and head pose (in contrast to Stable Diffusion Inpainting) and enables selective anonymization of specific facial regions.
For the quantitative comparisons against baseline methods in our paper, we selected:
For each subject, we created corresponding segmentation masks to selectively control the visibility of the eye and mouth areas if desired.
The list of selected test subjects and their corresponding segmentation masks are available for download at the Hugging Face Hub.