Research map / Label-efficient and robust learning / Self-supervised and contrastive
On Model and Data Scaling for Skeleton-based Self-Supervised Gait Recognition
Adrian Cosma, Andy Eduard Catruna, Emilian Radoi
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 |
|---|---|---|---|
| select top-quality skeleton sequences by confidence-based quality score instead of random subsets (ablation) | GaitPTv2 (mu=2) · controlled gait benchmarks | data scaling exponent beta: 0.078 -> 0.1319 | Improved |
| select top-quality skeleton sequences instead of random subsets (ablation) | GaitPTv2 (mu=2) · in-the-wild gait benchmarks | data scaling exponent beta: 0.099 -> 0.125 | Improved |
| scale pretraining data to the full 2.7M sequences | GaitPTv2 · controlled gait benchmarks | model scaling exponent alpha: 0.100 -> 0.164 | Improved |
| scale pretraining data from 43.4K to 2.78M sequences | GaitPTv2 · in-the-wild gait benchmarks | model scaling exponent alpha: 0.118 -> 0.112 | No clear effect |
| increase model size | GaitPTv2 · controlled and in-the-wild gait benchmarks | zero-shot accuracy | Improved |
| increase pretraining data to the largest scale | GaitPTv2 · gait benchmarks | zero-shot accuracy | Mixed |
| use hierarchical GaitPTv2 instead of GaitFormer under same parameter count | gait encoder · controlled gait recognition | accuracy | Improved |
| use hierarchical GaitPTv2 instead of GaitFormer | gait encoder · controlled gait recognition | compute efficiency (FLOPs): 2 -> 6.5 | Hurt |
| increase parameter count at fixed compute (IsoFLOPs) | GaitPTv2 and GaitFormer · controlled gait recognition | accuracy | Hurt |
| scale parameter count of GaitPTv2 vs GaitFormer using compute-efficient models | gait encoder · controlled gait recognition | accuracy | No clear effect |
| scale number of training sequences for GaitPTv2 vs GaitFormer | gait encoder · controlled gait recognition | accuracy | Improved |
Closest papers by meaning
- Gait Recognition in the Wild: A Large-Scale Benchmark and NAS-Based Baseline (TPAMI 2025 · similarity 0.91)
- Learning Gait Representation From Massive Unlabelled Walking Videos: A Benchmark (TPAMI 2023 · similarity 0.90)
- BiggerGait: Unlocking Gait Recognition with Layer-wise Representations from Large Vision Models (NeurIPS 2025 · similarity 0.87)
- SkeletonGait: Gait Recognition Using Skeleton Maps (AAAI 2024 · similarity 0.87)
- Learning Visual Prompt for Gait Recognition (CVPR 2024 · similarity 0.86)
- Exploring More from Multiple Gait Modalities for Human Identification (AAAI 2025 · similarity 0.86)
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.
