Research map / Label-efficient and robust learning / Self-supervised and contrastive
A Self-Supervised Gait Encoding Approach With Locality-Awareness for 3D Skeleton Based Person Re-Identification
Haocong Rao, Siqi Wang, Xiping Hu, Mingkui Tan, Yi Guo, Jun Sheng Cheng, Xinwang Liu, Bin Hu
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 decoder (self-supervised reconstruction) to the LSTM encoder (ablation) | GE · BIWI | Rank-1: 36.1 -> 41.5 | Improved |
| add reverse reconstruction as pretext task (ablation) | GE+GD · BIWI | Rank-1: 41.5 -> 46.7 | Improved |
| add locality-aware attention (LAS) (ablation) | GE+GD+Rev. Rec. · BIWI | Rank-1: 46.7 -> 57.7 | Improved |
| add locality-aware contrastive learning (CAGEs vs AGEs) (ablation) | full model · BIWI | Rank-1: 59.1 -> 62.9 | Improved |
| use reverse reconstruction instead of prediction as pretext (ablation) | self-supervised gait encoding · BIWI | Rank-1 (AP): 40.4 -> 62.9 | Improved |
| use reverse reconstruction instead of sorting as pretext (ablation) | self-supervised gait encoding · BIWI | Rank-1 (AP): 55.7 -> 62.9 | Improved |
| combine all three pretext tasks (Rev. Rec. Plus) vs Rev. Rec. alone (ablation) | CAGEs · KS20 | Rank-1 (AP): 86.9 -> 92.0 | Improved |
| combine all three pretext tasks vs Rev. Rec. alone (ablation) | CAGEs · IAS-A | Rank-1 (AP): 60.1 -> 59.1 | Hurt |
| use average prediction (AP) instead of sequence-level concatenation (SC) (ablation) | CAGEs Rev. Rec. · BIWI | Rank-1: 51.3 -> 62.9 | Improved |
| contrast non-adjacent sequences (interval 2) instead of adjacent (ablation) | LCL · BIWI | Rank-1: 62.9 -> 61.1 | Hurt |
| raise LCL temperature from 0.1 to 1 (ablation) | LCL · BIWI | Rank-1: 62.9 -> 60.6 | No clear effect |
| apply attention (BAS, MBAS, LAS) vs no attention (ablation) | reverse reconstruction model · multiple datasets | Rank-1: 1.0%-7.2% Rank-1 | Improved |
Closest papers by meaning
- Motif Guided Graph Transformers with Combinatorial Skeleton Prototype Learning for Skeleton-Based Person Re-Identification (AAAI 2025 · similarity 0.91)
- Hierarchical Skeleton Meta-Prototype Contrastive Learning with Hard Skeleton Mining for Unsupervised Person Re-identification (IJCV 2023 · similarity 0.88)
- On Model and Data Scaling for Skeleton-based Self-Supervised Gait Recognition (AAAI 2026 · similarity 0.88)
- Learning Gait Representation From Massive Unlabelled Walking Videos: A Benchmark (TPAMI 2023 · similarity 0.87)
- SkeletonGait: Gait Recognition Using Skeleton Maps (AAAI 2024 · similarity 0.87)
- Exploring More from Multiple Gait Modalities for Human Identification (AAAI 2025 · similarity 0.85)
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.
