Research map / Image generation / Autoregressive and latent generation
DiffCrossGait: Trajectory-Level Alignment for 2D-3D Cross-Modal Gait Recognition via Latent Diffusion
Zhiyang Lu, Ming Cheng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add trajectory-level shared-noise diffusion alignment training objective to the ResNet-9 baseline | DiffCrossGait · SUSTech1K 3D->2D | Rank-1: 54.7 -> 63.8 | Improved |
| add Phase1 identity anchoring alone (ablation) | Tri-Phase Alignment · SUSTech1K | Rank-1: 54.7 -> 60.6 | Improved |
| add Phase2 dynamics consistency on top of Phase1 (ablation) | Tri-Phase Alignment · SUSTech1K | Rank-1: 60.6 -> 63.1 | Improved |
| add Phase3 cross-modal recoverability on top of Phase1+2 (ablation) | Tri-Phase Alignment · SUSTech1K | Rank-1: 63.1 -> 63.8 | Improved |
| use shared Gaussian noise for both modalities (ablation) | DiffCrossGait · SUSTech1K | Rank-1 | Improved |
| add shared modal token (ablation) | DiffCrossGait denoiser · SUSTech1K | Rank-1 | Improved |
| add decoupled discriminative/generative heads (ablation) | DiffCrossGait · SUSTech1K | Rank-1 | Improved |
| add axial attention in bottleneck (ablation) | DiffCrossGait denoiser · SUSTech1K | Rank-1 | Improved |
| share one denoiser across modalities instead of independent U-Nets (ablation) | DiffCrossGait · SUSTech1K | Rank-1: 63.3 -> 63.8 | Improved |
| use heavyweight deep U-Net instead of lightweight denoiser (ablation) | DiffCrossGait · SUSTech1K | Rank-1: 63.8 -> 59.1 | Hurt |
| vary phase boundaries, diffusion length and loss weights (ablation) | DiffCrossGait · SUSTech1K | Rank-1: 58.1 to 58.7 across phase/length settings | No clear effect |
| apply diffusion objective to DEEN on SYSU-MM01 | DEEN · SYSU-MM01 all-search | Rank-1: 74.70 -> 75.19 | Improved |
Closest papers by meaning
- On Denoising Walking Videos for Gait Recognition (CVPR 2025 · similarity 0.87)
- Text-guided Feature Disentanglement for Cross-modal Gait Recognition (CVPR 2026 · similarity 0.86)
- Gait-X: Exploring X modality for Generalized Gait Recognition (ICCV 2025 · similarity 0.86)
- HybridGait: A Benchmark for Spatial-Temporal Cloth-Changing Gait Recognition with Hybrid Explorations (AAAI 2024 · similarity 0.85)
- Cross Flow: One-Step Generation Across Latent and Pixel Spaces (NeurIPS 2026 · similarity 0.84)
- Cross-Covariate Gait Recognition: A Benchmark (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
