Pedestrian Attribute Editing for Gait Recognition and Anonymization

TL;DR

GaitEditor is the first framework capable of editing multiple gait attributes while simultaneously benefiting gait recognition and gait anonymization, and is believed to be the first framework capable of editing multiple gait attributes while simultaneously benefiting gait recognition and gait anonymization.

Abstract

As a kind of biometrics, the gait information of pedestrians has attracted widespread attention from both industry and academia since it can be acquired from long distances without the cooperation of targets. In recent literature, this line of research has brought exciting chances along with alarming challenges: On the positive side, gait recognition used for security applications such as suspect retrieval and safety checks is becoming more and more promising. On the negative side, the misuse of gait information may lead to privacy concerns, as lawbreakers can track subjects of interest using gait characteristics even under face-masked and clothes-changed scenarios. To handle this double-edged sword, we propose a gait attribute editing framework termed GaitEditor. It can perform various degrees of attribute edits on real gait sequences while maintaining the visual authenticity, respectively used for gait data augmentation and de-identification, thereby adaptively enhancing or degrading gait recognition performance according to users' intentions. Experimentally, we conduct a comprehensive evaluation under both gait recognition and anonymization protocols on three widely used gait benchmarks. Numerous results illustrate that the adaptable utilization of GaitEditor efficiently improves gait recognition performance and generates vivid visualizations with de-identification to protect human privacy. To the best of our knowledge, GaitEditor is the first framework capable of editing multiple gait attributes while simultaneously benefiting gait recognition and gait anonymization. The source code of GaitEditor will be available at https://github.com/ShiqiYu/OpenGait.

Figures (9)

• Figure 1: We present GaitEditor (sub-figure (a)) to enable attribute-conditioned semantic edits on real gait sequences (sub-figure (b)), thereby boosting the performance of gait recognition, all while mitigating privacy concerns (sub-figure (c)). For GaitEditor, we project the real gait data into the built latent space by well-designed encoder $\mathbf{E}$, manipulate latent codes in various semantic directions and project the altered ones to the image domain via the generator $\mathbf{G}$.
• Figure 2: GaitEditor's overall training architecture. The training process involves two stages. (a) An image-based GAN model is trained on an unlabeled silhouette dataset to establish a latent space. (b) GaitEditor involves two branches, inversion and viewpoint translation. The first branch involves projecting the source gait sequence into the established latent space, as well as the other branch focuses on altering the viewpoint from the target sample to that of the source sample. To ensure that all the relevant information in gait sequences is preserved, four losses are designed, and they are identity ($\mathcal{L}_{id}$), viewpoint ($\mathcal{L}_{view}$), video adversarial ($\mathcal{L}_{adv}$), and reconstruction ($\mathcal{L}_{rec}$) losses.
• Figure 3: Interpolation results for our build latent space. Each row sequence showcases the stages between two unique random noises. The first and last images in every row are the synthetic visualizations generated from these individual noises.
• Figure 4: The architecture of Attribute-Identity encoder.
• Figure 5: The schematic of editing non-viewpoint attributes.