Research map / Video / Action recognition
LAC - Latent Action Composition for Skeleton-based Action Segmentation
Di Yang, Yaohui Wang, Antitza Dantcheva, Quan Kong, Lorenzo Garattoni, Gianpiero Francesca, Francois Bremond
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 motion composition from two skeleton sequences in pre-training (ablation) | UNIK contrastive pre-training · TSU CS | frame mAP (%): 29.8 -> 33.8 | Improved |
| compose three motions instead of two (ablation) | LAC · TSU CS | frame mAP (%): 33.8 -> 32.1 | Hurt |
| add frame-level contrastive learning (sample every 4 frames) (ablation) | LAC · TSU CS | frame mAP (%): 33.8 -> 34.1 | Improved |
| use temporal sample rate 8 instead of 4 (ablation) | LAC frame-level contrast · TSU CS | frame mAP (%): 34.1 -> 33.7 | Hurt |
| motion dictionary size J=128 (of 160 directions) (ablation) | LAC generator · Mixamo retargeting | MSE: 1.23 -> 0.82 | Improved |
| self-supervised pre-training on Posetics versus random init (fine-tuning) | UNIK skeleton encoder · TSU CS | frame mAP (%): 28.2 -> 34.1 | Improved |
| self-supervised pre-training with 5% labeled data | UNIK skeleton encoder · TSU CS | frame mAP (%): 8.5 -> 25.2 | Improved |
| self-supervised instead of supervised pre-training (fine-tuning) | LAC · TSU CS | frame mAP (%): 36.8 -> 34.1 | Hurt |
| pre-train on small target dataset only instead of Posetics | LAC · TSU CS | frame mAP (%): 34.1 -> 33.2 | Hurt |
| LAC on PKU-MMD lab data without composable actions | LAC · PKU-MMD CS IoU 0.1 | event mAP (%): 92.2 vs 92.6 | No clear effect |
Closest papers by meaning
- Skeleton Motion Words for Unsupervised Skeleton-Based Temporal Action Segmentation (ICCV 2025 · similarity 0.92)
- Stitch, Contrast, and Segment: Learning a Human Action Segmentation Model Using Trimmed Skeleton Videos (AAAI 2025 · similarity 0.91)
- HaLP: Hallucinating Latent Positives for Skeleton-Based Self-Supervised Learning of Actions (CVPR 2023 · similarity 0.90)
- Recovering Complete Actions for Cross-dataset Skeleton Action Recognition (NeurIPS 2024 · similarity 0.89)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.89)
- Heterogeneous Skeleton-Based Action Representation Learning (CVPR 2025 · 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.
