Research map / Video / Action recognition
Path-JEPA: Path Signature Based Predictive Learning for Skeleton Action Recognition
Ashutosh Singh, Ashish Singh
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 local window signatures to global signatures (ablation) | Path-JEPA joint tokens · NTU60 X-Sub (linear eval) | accuracy (%): 84.4 -> 85.8 | Improved |
| add edge tokens (ablation) | Path-JEPA · NTU60 X-Sub (linear eval) | accuracy (%): 85.8 -> 86.4 | Improved |
| add chain tokens (ablation) | Path-JEPA · NTU60 X-Sub (linear eval) | accuracy (%): 86.4 -> 87.6 | Improved |
| add global signature to local-only tokens (ablation) | Path-JEPA · NTU60 X-Sub (linear eval) | accuracy (%): 83.3 -> 85.8 | Improved |
| signature-augmented masking instead of random joint masking (ablation) | Path-JEPA · NTU60 X-Sub (linear eval) | accuracy (%): 84.1 -> 86.3 | Improved |
| add kinematic subtree expansion to masking (ablation) | Path-JEPA · NTU60 X-Sub (linear eval) | accuracy (%): 86.3 -> 87.6 | Improved |
| vary mask ratio from default (about 20-25%) to 10% or 40% (ablation) | Path-JEPA · NTU60 X-Sub (linear eval) | accuracy (%): 87.6 -> 85.1 (10%), 84.9 (40%) | Hurt |
| use path-signature targets instead of S-JEPA latent targets (frame-rate robustness) | Path-JEPA · NTU60 X-Sub (15fps resampled) | accuracy (%): 82.4 -> 87.1 | Improved |
| Path-JEPA pretraining vs S-JEPA, 1% labels | Path-JEPA · NTU60 X-Sub (semi-supervised, 1% labels) | accuracy (%): 67.5 -> 72.2 | Improved |
| Path-JEPA pretraining vs S-JEPA, 60% labels | Path-JEPA · NTU60 X-Sub (semi-supervised, 60% labels) | accuracy (%): 92.1 -> 92.1 | No clear effect |
| Path-JEPA vs S-JEPA, transfer to PKU-II | Path-JEPA · PKU-II transfer (pretrain NTU60) | accuracy (%): 71.4 -> 72.6 | Improved |
Closest papers by meaning
- S-JEPA: A Joint Embedding Predictive Architecture for Skeletal Action Recognition (ECCV 2024 · similarity 0.90)
- Self-Supervised Skeleton-Based Action Representation Learning: A Benchmark and Beyond (IJCV 2026 · similarity 0.90)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · similarity 0.88)
- Neuron: Learning Context-Aware Evolving Representations for Zero-Shot Skeleton Action Recognition (CVPR 2025 · similarity 0.88)
- SAMoR: Motion Modelling for Articulated Objects of Any Skeleton and Topology (NeurIPS 2026 · similarity 0.87)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.87)
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.
