Research map / Video / Action recognition
Cooperative Training of Deep Aggregation Networks for RGB-D Action Recognition
Pichao Wang, Wanqing Li, Jun Wan, Philip Ogunbona, Xinwang Liu
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 |
|---|---|---|---|
| Train one network cooperatively on RGB and depth with joint softmax and intra-/cross-modality ranking losses instead of separate conventional ConvNets | RGB-D dynamic-image action recognition · ChaLearn LAP IsoGD validation | accuracy: 35.65% -> 44.80% | Improved |
| Train one network cooperatively on RGB and depth with joint softmax and intra-/cross-modality ranking losses instead of separate conventional ConvNets | RGB-D dynamic-image action recognition · NTU RGB+D cross-subject | accuracy: 84.99% -> 86.42% | Improved |
| Train one network cooperatively on RGB and depth with joint softmax and intra-/cross-modality ranking losses instead of separate conventional ConvNets | RGB-D dynamic-image action recognition · NTU RGB+D cross-view | accuracy: 87.51% -> 89.08% | Improved |
| Train one network cooperatively on RGB and depth with joint softmax and intra-/cross-modality ranking losses instead of separate conventional ConvNets | RGB-D dynamic-image action recognition · SYSU 3D HOI | accuracy: 97.92% -> 98.33% | Improved |
| Use all four forward and backward RGB/depth dynamic-image channels rather than only one or two channels (ablation) | c-ConvNet · ChaLearn LAP IsoGD validation | accuracy: 41.01% -> 44.80% | Improved |
| Use all four forward and backward RGB/depth dynamic-image channels rather than only one or two channels (ablation) | c-ConvNet · NTU RGB+D cross-view | accuracy: 85.21% -> 89.08% | Improved |
| Replace product score fusion with max or average score fusion (ablation) | Four-channel c-ConvNet predictions · ChaLearn LAP IsoGD | accuracy: 44.80% vs. 42.01% / 43.48% | Hurt |
| Replace product score fusion with max or average score fusion (ablation) | Four-channel c-ConvNet predictions · NTU RGB+D cross-subject | accuracy: 86.42% vs. 84.69% / 85.86% | Hurt |
| Replace product score fusion with max or average score fusion (ablation) | Four-channel c-ConvNet predictions · SYSU 3D HOI | accuracy: 98.33% vs. 97.08% / 97.92% | Hurt |
| Increase the cross-modality triplet-loss weight λ (ablation) | c-ConvNet DDIf + VDIf · ChaLearn LAP IsoGD | accuracy: 39.68 -> 39.51 -> 39.61 -> 39.71 -> 41.01 -> 40.13 | Mixed |
| Increase the cross-modality triplet-loss weight λ (ablation) | c-ConvNet DDIf + VDIf · NTU RGB+D cross-subject | accuracy: 80.36 -> 81.15 -> 82.64 -> 80.18 -> 80.06 -> 80.11 | Mixed |
| Use bidirectional rather than forward-only dynamic images (ablation) | RGB-D c-ConvNet · ChaLearn LAP IsoGD validation | accuracy | Improved |
Closest papers by meaning
- Deep Multimodal Feature Analysis for Action Recognition in RGB+D Videos (TPAMI 2017 · similarity 0.88)
- MMNet: A Model-Based Multimodal Network for Human Action Recognition in RGB-D Videos (TPAMI 2022 · similarity 0.87)
- Deep Bilinear Learning for RGB-D Action Recognition (ECCV 2018 · similarity 0.87)
- Modality Distillation with Multiple Stream Networks for Action Recognition (ECCV 2018 · similarity 0.86)
- DVANet: Disentangling View and Action Features for Multi-View Action Recognition (AAAI 2024 · similarity 0.86)
- Are Spatial-Temporal Graph Convolution Networks for Human Action Recognition Over-Parameterized? (CVPR 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.
