Research map / Video / Action recognition
Learning by Aligning 2D Skeleton Sequences and Multi-Modality Fusion
Quoc-Huy Tran, Muhammad Ahmed, Murad Popattia, Muhammad Hassan Ahmed, Andrey Konin, Zeeshan Zia
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 spatial transformer to the heatmap video transformer (ablation) | LA2DS · H2O | phase progression: 0.9039 -> 0.9280 | Improved |
| Add temporal transformer (ablation) | LA2DS · H2O | phase classification accuracy: 64.40 -> 70.12 | Improved |
| Use joint+limb heatmaps instead of joints only (ablation) | LA2DS · H2O | phase classification accuracy: 66.09 -> 70.12 | Improved |
| Add all four 2D heatmap augmentations versus none (ablation) | LA2DS · H2O | phase classification accuracy: 63.12 -> 70.12 | Improved |
| Add temporal augmentation (ablation) | LA2DS · H2O | phase classification accuracy: 64.82 -> 70.12 | Improved |
| Use temporally smooth correlated noise for position and orientation augmentation | LA2DS · H2O | accuracy | No clear effect |
| Use 2D heatmaps instead of 3D coordinates under missing keypoints (p=100%) | LA2DS vs CASA · H2O | phase classification accuracy: 60.93 -> 68.94 | Improved |
| Use 2D heatmaps instead of 3D coordinates under noisy keypoints | LA2DS vs CASA · H2O | phase classification accuracy: 63.69 -> 68.23 | Improved |
| Fuse RGB with pose heatmaps (late fusion) | LA2DS · Penn Action | phase classification accuracy: 92.63 -> 93.57 | Improved |
| Use video transformer ViViT on RGB inside LAV | LAV · Penn Action | phase classification accuracy: 83.56 -> 70.66 | Hurt |
| Use STGCN encoder in CASA | CASA · Penn Action | phase classification accuracy: 88.55 -> 87.45 | Hurt |
| Use skeleton-only input versus RGB methods with few labels | LA2DS · IKEA ASM | phase classification accuracy: 26.43 vs 29.78 (LAV) | Mixed |
Closest papers by meaning
- Context-Aware Sequence Alignment Using 4D Skeletal Augmentation (CVPR 2022 · similarity 0.89)
- Self-Supervised Skeleton-Based Action Representation Learning: A Benchmark and Beyond (IJCV 2026 · similarity 0.88)
- Modeling the Relative Visual Tempo for Self-supervised Skeleton-based Action Recognition (ICCV 2023 · similarity 0.88)
- Self-Supervised 3D Skeleton Action Representation Learning With Motion Consistency and Continuity (ICCV 2021 · similarity 0.87)
- Meet JEANIE: A Similarity Measure for 3D Skeleton Sequences via Temporal-Viewpoint Alignment (IJCV 2024 · similarity 0.87)
- Stitch, Contrast, and Segment: Learning a Human Action Segmentation Model Using Trimmed Skeleton Videos (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.
