Research map / Detection and segmentation / 3D detection and driving
Camera-LiDAR Cross-modality Gait Recognition
Wenxuan Guo, Yingping Liang, Zhiyu Pan, Ziheng Xi, Jianjiang Feng, Jie Zhou
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 |
|---|---|---|---|
| use two-stream network with modality-specific shallow layers instead of one-stream shared network (ablation) | cross-modality gait network · SUSTech1K LiDAR to Camera | Rank-1: 2.60 -> 30.39 | Improved |
| add horizontal pooling to two-stream ResNet (ablation) | two-stream ResNet · SUSTech1K LiDAR to Camera | Rank-1: 30.39 -> 45.41 | Improved |
| add contrastive silhouette-point pre-training (CSPP) on top of HP (ablation) | two-stream ResNet+HP · SUSTech1K LiDAR to Camera | Rank-1: 45.41 -> 53.29 | Improved |
| add CSPP without horizontal pooling (ablation) | two-stream ResNet · SUSTech1K LiDAR to Camera | Rank-1: 30.39 -> 32.18 | Improved |
| project point clouds to depth images instead of using raw points with point encoders (ablation) | cross-modality gait network · SUSTech1K LiDAR to Camera | Rank-1: 13.25 -> 30.39 | Improved |
| use depth images instead of LiDAR silhouettes lacking 3D geometry (ablation) | CL-Gait · SUSTech1K | average Rank-1: 40.59% -> 54.21% | Improved |
| pre-train on generated pseudo pairs from LIP (50K) versus real SUSTech1K pairs (120K) (ablation) | CSPP · SUSTech1K LiDAR to Camera | Rank-1: 51.60 -> 51.00 | No clear effect |
| scale pseudo-data pre-training to HITSZ-VCM (770K) versus real data (ablation) | CSPP · SUSTech1K | Rank-1 L to C / C to L: L to C 51.60 -> 53.29; C to L 56.12 -> 55.12 | Mixed |
| replace temporal pooling with LSTM, Bi-LSTM or Transformer (ablation) | CL-Gait · SUSTech1K LiDAR to Camera | Rank-1: 53.29 -> 37.40 (LSTM), 39.63 (Bi-LSTM), 44.80 (Transformer) | Hurt |
| use asymmetric FC layer for modality adaptation instead of shallow modality-specific layers (ablation) | cross-modality gait network · SUSTech1K LiDAR to Camera | Rank-1: 3.08 vs 30.39 for two-stream | Hurt |
Closest papers by meaning
- LidarGait: Benchmarking 3D Gait Recognition With Point Clouds (CVPR 2023 · similarity 0.90)
- Text-guided Feature Disentanglement for Cross-modal Gait Recognition (CVPR 2026 · similarity 0.90)
- MMGait: Towards Multi-Modal Gait Recognition (CVPR 2026 · similarity 0.89)
- LidarGait++: Learning Local Features and Size Awareness from LiDAR Point Clouds for 3D Gait Recognition (CVPR 2025 · similarity 0.85)
- Cross-Covariate Gait Recognition: A Benchmark (AAAI 2024 · similarity 0.84)
- Exploring More from Multiple Gait Modalities for Human Identification (AAAI 2025 · 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.
