Research map / Video / Action recognition
Self-Supervised 3D Action Representation Learning With Skeleton Cloud Colorization
Siyuan Yang, Jun Liu, Shijian Lu, M.H. Er, Yongjian Hu, Alex Chichung Kot
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 |
|---|---|---|---|
| pretrain with skeleton cloud colorization instead of random encoder (ablation) | linear classifier · NTU RGB+D C-Subject (unsupervised) | accuracy: 61.8 -> 75.2 | Improved |
| replace plain masked autoencoder pretraining with colorization repainting (ablation) | 3s model · NTU RGB+D C-Subject (unsupervised) | accuracy: 65.9 -> 75.2 | Improved |
| add masked skeleton cloud repainting (ablation) | 3s-Colorization · NTU RGB+D C-Subject (unsupervised) | accuracy: 75.2 -> 77.2 | Improved |
| add coarse-fine alignment on top of masking (ablation) | 3s-Masked Colorization · NTU RGB+D C-Subject (unsupervised) | accuracy: 77.2 -> 79.1 | Improved |
| add coarse-fine alignment to spatial stream (ablation) | SS Masked Colorization · NTU RGB+D C-Subject (unsupervised) SS | accuracy: 69.6 -> 72.5 | Improved |
| add person-level colorization stream to TS+SS (ablation) | TS+SS Colorization · NTU RGB+D C-Subject (unsupervised) | accuracy: 74.6 -> 75.2 | Improved |
| use segment masking (15 frames) instead of random masking 0.75 (ablation) | TS stream · NTU RGB+D C-Subject (unsupervised) TS | accuracy: 72.8 -> 74.0 | Improved |
| use 10-joint masking instead of random masking 0.75 (ablation) | SS stream · NTU RGB+D C-Subject (unsupervised) SS | accuracy: 71.7 -> 72.8 | Improved |
| use 5-frame segments for coarse temporal colorization instead of 2 (ablation) | TS stream · NTU RGB+D C-Subject (unsupervised) | accuracy: 72.2 -> 73.2 | Improved |
| use 10 body parts instead of 6 for coarse spatial colorization (ablation) | SS stream · NTU RGB+D C-Subject (unsupervised) | accuracy: 71.3 -> 72.5 | Improved |
| add person-level colorization masking (ablation) | PS stream · NTU RGB+D C-Subject (unsupervised) PS | accuracy: 64.2 -> 67.9 | Improved |
| coarse-fine masked colorization vs prior colorization | 3s-Colorization · Transfer NTU RGB+D to PKU-MMD II | accuracy: 51.0 -> 58.1 | Improved |
Closest papers by meaning
- Skeleton Cloud Colorization for Unsupervised 3D Action Representation Learning (ICCV 2021 · similarity 0.97)
- Self-Supervised Skeleton-Based Action Representation Learning: A Benchmark and Beyond (IJCV 2026 · similarity 0.91)
- Self-Supervised 3D Skeleton Action Representation Learning With Motion Consistency and Continuity (ICCV 2021 · similarity 0.90)
- SkeletonMAE: Graph-based Masked Autoencoder for Skeleton Sequence Pre-training (ICCV 2023 · similarity 0.90)
- Revealing Key Details to See Differences: A Novel Prototypical Perspective for Skeleton-based Action Recognition (CVPR 2025 · similarity 0.88)
- Heterogeneous Skeleton-Based Action Representation Learning (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.
