Research map / Label-efficient and robust learning / Semi-supervised learning
Adversarial Self-Supervised Learning for Semi-Supervised 3D Action Recognition
Chenyang Si, Xuecheng Nie, Wei Wang, Liang Wang, Tieniu Tan, Jiashi Feng
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 neighborhood consistency regularization to Sup. (ablation) | Sup. + Inp. · NTU RGB+D CS (5% labels) | accuracy: 48.4 -> 55.2 | Improved |
| add neighborhood consistency instead of inpainting (ablation) | Supervised-only · NTU RGB+D CS (5% labels) | accuracy: 48.4 -> 52.1 | Improved |
| add neighborhood consistency instead of inpainting (ablation) | Supervised-only · N-UCLA (15% labels) | accuracy: 46.6 -> 60.0 | Improved |
| add skeleton inpainting to neighborhood consistency (ablation) | Sup. + Nei. · N-UCLA (15% labels) | accuracy: 60.0 -> 66.4 | Improved |
| add adversarial regularization to Sup. + Inp. + Nei. (ablation) | ASSL · N-UCLA (15% labels) | accuracy: 66.4 -> 74.8 | Improved |
| add adversarial regularization to Sup. + Inp. + Nei. (ablation) | ASSL · NTU RGB+D CS (5% labels) | accuracy: 55.2 -> 57.3 | Improved |
| add adversarial regularization to Sup. + Nei. (ablation) | Sup. + Nei. · NTU RGB+D CS (5% labels) | accuracy: 52.1 -> 53.4 | Improved |
| add adversarial regularization to Sup. + Inp. (ablation) | Sup. + Inp. · N-UCLA (15% labels) | accuracy: 46.6 -> 52.4 | Improved |
| add skeleton inpainting self-supervision (S4L) to supervised training | Supervised-only · N-UCLA (5% labels) | accuracy: 34.1 -> 35.3 | No clear effect |
| add skeleton inpainting self-supervision (S4L) to supervised training | Supervised-only · NTU RGB+D CS (5% labels) | accuracy: 47.2 -> 48.4 | No clear effect |
| replace VAT + EntMin with ASSL | VAT + EntMin · NTU RGB+D CS (5% labels) | accuracy: 51.7 -> 57.3 | Improved |
| replace VAT + EntMin with ASSL | VAT + EntMin · NTU RGB+D CS (40% labels) | accuracy: 70.8 -> 72.3 | Improved |
Closest papers by meaning
- Improving 3D Semi-supervised Learning by Effectively Utilizing All Unlabelled Data (ECCV 2024 · similarity 0.86)
- Semi-Supervised Active Learning for Semi-Supervised Models: Exploit Adversarial Examples With Graph-Based Virtual Labels (ICCV 2021 · similarity 0.86)
- Motion-Augmented Self-Training for Video Recognition at Smaller Scale (ICCV 2021 · similarity 0.86)
- Unsupervised Learning of View-invariant Action Representations (NeurIPS 2018 · similarity 0.85)
- Leveraging Temporal Cues for Semi-Supervised Multi-View 3D Object Detection (CVPR 2025 · similarity 0.84)
- TRiCo: Triadic Game-Theoretic Co-Training for Robust Semi-Supervised Learning (NeurIPS 2025 · similarity 0.84)
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.
