Reproduce “GaitCrafter: Diffusion Model for Biometric Preserving Gait Synthesis” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sirshapan Mitra, Yogesh S. Rawat
A video diffusion model, GaitCrafter, synthesizes realistic, temporally consistent, and identity-preserving gait silhouettes, including novel identities and controllable covariate variations like viewpoint and clothing. This framework improves downstream gait recognition performance on the CASIA-B dataset, boosting overall average accuracy by 0.9% and clothing condition performance by 2% when augmenting with novel synthetic identities.
Key insights
- A pixel-based video diffusion model can effectively generate high-fidelity and identity-preserving silhouettes, overcoming challenges of binary data distribution and maintaining temporal consistency.
- Expanding the identity space with synthetic novel individuals provides greater benefits for improving gait recognition model generalization in open-set or low-identity scenarios than simply augmenting samples for existing identities.
- Training gait recognition models exclusively on synthetic novel identities can achieve performance comparable to models trained on real data, offering a viable solution for privacy-constrained or data-scarce environments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.