Research map / Label-efficient and robust learning / Domain adaptation and generalization
Collaborating Domain-Shared and Target-Specific Feature Clustering for Cross-Domain 3D Action Recognition
Qinying Liu, Zilei Wang
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 online clustering module to base modules (ablation) | CoDT-B0 · N->S (NTU-60 to Skeletics) | ACC (%): 17.9 -> 22.0 | Improved |
| add online clustering module to base modules (ablation) | CoDT-B1 · N+->P (NTU-60+ to PKUMMD) | ACC (%): 42.7 -> 52.6 | Improved |
| add collaborative clustering module on top of OCM (ablation) | CoDT-B0 · N->S (NTU-60 to Skeletics) | ACC (%): 22.0 -> 25.0 | Improved |
| add collaborative clustering module on top of OCM (ablation) | CoDT-B1 · N+->P (NTU-60+ to PKUMMD) | ACC (%): 52.6 -> 67.8 | Improved |
| add uniform (optimal transport) constraint to pseudo labels (ablation) | CoDT-B0 · S->P (Skeletics to PKUMMD) | ACC (%): 31.2 -> 55.7 | Improved |
| add loss regularization L to pseudo-label clustering (ablation) | CoDT-B0 · S->P (Skeletics to PKUMMD) | ACC (%): 9.0 -> 31.2 | Improved |
| add uniform constraint when CCM is used (ablation) | CoDT-B0 · S->P (Skeletics to PKUMMD) | ACC (%): 48.4 -> 59.5 | Improved |
| add decoder reconstruction to domain-shared branch (ablation) | BM-B0 · N->S (NTU-60 to Skeletics) | ACC (%): 16.6 -> 17.9 | Improved |
| add decoder reconstruction to domain-shared branch (ablation) | BM-B0 · S->P (Skeletics to PKUMMD) | ACC (%): 38.5 -> 40.0 | Improved |
| combine B0 and B1 via multi-task learning instead of single branch (ablation) | BM · N+->P (NTU-60+ to PKUMMD) | ACC (%): 54.8 -> 58.3 | Improved |
| use pair-wise labels from pseudo-label comparison and prediction similarity (CCM) instead of feature-similarity pair labels (CCM-FP) (ablation) | CCM · N+->P (NTU-60+ to PKUMMD) | ACC (%): 63.9/55.3 -> 68.1/67.8 | Improved |
| use prediction-similarity optimization (CCM) instead of feature-similarity optimization (CCM-PF) (ablation) | CCM · N+->P (NTU-60+ to PKUMMD) | ACC (%): 64.7/55.9 -> 68.1/67.8 | Improved |
Closest papers by meaning
- CMD: Self-Supervised 3D Action Representation Learning with Cross-Modal Mutual Distillation (ECCV 2022 · similarity 0.87)
- $$\hbox {I}^2$$MD: 3D Action Representation Learning with Inter- and Intra-Modal Mutual Distillation (IJCV 2025 · similarity 0.85)
- Interact Before Align: Leveraging Cross-Modal Knowledge for Domain Adaptive Action Recognition (CVPR 2022 · similarity 0.85)
- 3D Human Action Representation Learning via Cross-View Consistency Pursuit (CVPR 2021 · similarity 0.85)
- CDFSL-V: Cross-Domain Few-Shot Learning for Videos (ICCV 2023 · similarity 0.84)
- Recovering Complete Actions for Cross-dataset Skeleton Action Recognition (NeurIPS 2024 · similarity 0.83)
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.
