GaitProtector: Impersonation-Driven Gait De-Identification via Training-Free Diffusion Latent Optimization

2026-05-12

AI Summary

This work addresses the challenge of preserving both identity privacy and spatiotemporal structural fidelity in gait de-identification. We propose a training-free diffusion latent-space optimization framework that, for the first time, leverages a pre-trained 3D video diffusion model without fine-tuning to perform silhouette-based gait de-identification. Guided by diffusion priors, our method jointly optimizes two objectives—obscuring the source identity while imitating a target identity—to generate privacy-preserving gait sequences with coherent structure. The approach combines latent inversion with differentiable adversarial objectives through iterative optimization, eliminating the need to retrain the generator. Experiments demonstrate that on CASIA-B, the black-box imitation attack achieves a 56.7% success rate, reducing Rank-1 identification accuracy from 89.6% to 15.0%; on Scoliosis1K, diagnostic accuracy only slightly drops from 91.4% to 74.2%, with high visual and temporal quality maintained.

Abstract

Conventional gait de-identification methods often encounter an inherent trade-off: they either provide insufficient identity suppression or introduce spatiotemporal distortions that impede structure-sensitive downstream applications. We propose GaitProtector, an impersonation-driven gait de-identification framework that formulates privacy protection as a unified objective with two tightly coupled components: (i) obfuscation, which repels the protected gait from the source identity, and (ii) impersonation, which attracts it toward a selected target identity. The target identity serves as a semantic anchor that biases optimization toward structurally plausible gait patterns under the pretrained diffusion prior, helping preserve dominant body shape and motion dynamics. We instantiate this idea through a training-free diffusion latent optimization pipeline. Instead of retraining a generator for each dataset, we invert each input silhouette sequence into the latent trajectory of a pretrained 3D video diffusion model and iteratively optimize latent codes with a differentiable adversarial objective to synthesize protected gaits. Experiments on the CASIA-B dataset show that GaitProtector achieves a 56.7% impersonation success rate under black-box gait recognition and reduces Rank-1 identification accuracy from 89.6% to 15.0%, while maintaining favorable visual and temporal quality. We further evaluate downstream utility on the Scoliosis1K dataset, where diagnostic accuracy decreases only from 91.4% to 74.2%. To the best of our knowledge, this work is the first to leverage pretrained 3D diffusion priors in a training-free manner for silhouette-based gait de-identification.

Problem

• gait de-identification
• identity suppression
• spatiotemporal distortion
• privacy protection
• downstream applications

Innovation

• gait de-identification
• diffusion latent optimization
• training-free
• impersonation-driven
• 3D video diffusion