Research map / 3D vision / Human pose estimation
Global-Local Motion Transformer for Unsupervised Skeleton-Based Action Learning
Boeun Kim, Hyung Jin Chang, Jungho Kim, Jin Young Choi
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 |
|---|---|---|---|
| disentangle local motion from the pose sequence (ablation) | GL-Transformer · NTU-60 xsub | accuracy: 71.1 -> 74.2 | Improved |
| disentangle local motion from the pose sequence (ablation) | GL-Transformer · NTU-60 xview | accuracy: 73.5 -> 81.9 | Improved |
| add global translational motion alongside local motion (GLA) (ablation) | GL-Transformer · NTU-60 xsub | accuracy: 74.2 -> 75.4 | Improved |
| add displacement magnitude prediction loss (ablation) | GL-Transformer · NTU-60 xsub | accuracy: 75.4 -> 75.7 | Improved |
| add frame intervals {1,5,10} in MPDP (ablation) | GL-Transformer · NTU-60 xsub | accuracy: 75.7 -> 76.3 | Improved |
| add interval 15 to {1,5,10} (ablation) | GL-Transformer · NTU-60 xsub | accuracy: 76.3 -> 75.7 | Hurt |
| add person-to-person attention (ablation) | GL-Transformer · NTU-120 xsub | accuracy: 64.9 -> 66.0 | Improved |
| add person-to-person attention (two-person categories) (ablation) | GL-Transformer · NTU-120 xsub | accuracy: 71.6 -> 73.5 | Improved |
| use trainable embedding instead of fixed sinusoidal (ablation) | GL-Transformer · NTU-60 xsub | accuracy: 75.5 -> 76.0 | Improved |
| use tight trainable embedding instead of trainable (ablation) | GL-Transformer · NTU-60 xsub | accuracy: 76.0 -> 76.3 | Improved |
| GL-Transformer instead of H-transformer | GL-Transformer vs H-transformer · NTU-60 | accuracy: +7.0% xsub, +11.0% xview | Improved |
Closest papers by meaning
- SkateFormer: Skeletal-Temporal Transformer for Human Action Recognition (ECCV 2024 · similarity 0.90)
- Unsupervised Representation Learning With Long-Term Dynamics for Skeleton Based Action Recognition (AAAI 2018 · similarity 0.89)
- Foundation Model for Skeleton-Based Human Action Understanding (TPAMI 2025 · similarity 0.89)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · similarity 0.88)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.88)
- SAMoR: Motion Modelling for Articulated Objects of Any Skeleton and Topology (NeurIPS 2026 · similarity 0.88)
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.
