ReGenHuman: Re-Generating Human Appearances for Realistic Full-Body Video Anonymization
Researchers have developed ReGenHuman, a novel pipeline for anonymizing full-body human appearances in videos. Unlike previous methods that blur or redact, ReGenHuman synthesizes entirely new human regions using identity-free structural cues like pose, segmentation, and depth. This approach aims to maintain realism and temporal consistency while ensuring anonymity, outperforming existing baselines in privacy, quality, and utility for downstream tasks such as video question answering. AI
IMPACT Enables more realistic and temporally consistent video anonymization for privacy-preserving applications.