ReGenHuman: Re-Generating Human Appearances for Realistic Full-Body Video Anonymization

2026-06-12

AI Summary

This work addresses the challenge of simultaneously achieving photorealism, temporal consistency, and strong privacy protection in human-centric video anonymization. To overcome the limitations of editing-based approaches, the authors propose a novel “regenerate rather than edit” paradigm that fully reconstructs human regions from structured cues. Leveraging 2D pose, semantic segmentation, and monocular depth as structural conditions, they fine-tune a video-to-video diffusion model and introduce the first end-to-end full-body anonymization pipeline. The method employs dual conditioning streams—StructAll and StructHuman—to enable identity-agnostic, high-fidelity reconstruction. Experiments demonstrate that the proposed approach significantly outperforms existing techniques in privacy preservation, visual quality, and temporal coherence, while maintaining excellent utility for downstream tasks such as video question answering, thereby achieving the first effective co-optimization of all three objectives.

Abstract

Anonymizing human-centric video data is an understudied problem. Prior anonymization techniques either blur or redact pixels at the cost of realism and downstream utility, or generate frame-by-frame at the cost of temporal coherence. We introduce ReGenHuman, the first full-body video anonymization pipeline that is simultaneously realistic, temporally consistent, and anonymous by construction. Contrary to past approaches which redact or edit the inputs directly, we propose a regenerate, don't edit paradigm. Our approach composites 2D pose, segmentation, and monocular depth into two complementary conditioning streams - StructAll and StructHuman, which are used to fine-tune a video-to-video diffusion backbone on in-the-wild human videos, synthesizing the human regions entirely from identity-free structural cues. We evaluate our model on privacy, quality, and utility, and show that our ReGenHuman achieves the best tradeoff across all three axes against current baselines. We further show that our anonymized videos remain effective for downstream tasks, including video question answering.

Problem

• video anonymization
• human appearance
• temporal coherence
• realism
• privacy

Innovation

• video anonymization
• diffusion models
• temporal coherence
• identity-free synthesis
• human appearance regeneration