Research map / Vision-language and multimodal / Cross-modal learning
$$\hbox {I}^2$$MD: 3D Action Representation Learning with Inter- and Intra-Modal Mutual Distillation
Yunyao Mao, Jiajun Deng, Wengang Zhou, Zhenbo Lu, Wanli Ouyang, Houqiang Li
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 CMD (cross-modal mutual distillation) (ablation) | MoCo v2 baseline plus IMD · NTU-60 x-sub linear | accuracy: 81.9 -> 83.4 | Improved |
| add IMD (intra-modal mutual distillation) (ablation) | MoCo v2 baseline plus CMD · NTU-60 x-sub linear | accuracy: 83.0 -> 83.4 | Improved |
| add CMD and IMD to baseline (ablation) | MoCo v2 · NTU-60 x-sub KNN | accuracy: 65.0 -> 75.9 | Improved |
| add CMD to I2MD (ablation) | I2MD · NTU-120 x-sub KNN | accuracy: 61.2 -> 62.0 | Improved |
| add IMD to I2MD (ablation) | I2MD · NTU-120 x-sub KNN | accuracy: 59.2 -> 62.0 | Improved |
| add bidirectional distillation between joint and motion (ablation) | CMD · NTU-60 x-sub linear (joint) | accuracy: 78.9 -> 79.8 | Improved |
| add motion-to-bone and all-pair bidirectional distillation (ablation) | CMD · NTU-60 x-sub KNN (joint) | accuracy: 63.4 -> 70.6 | Improved |
| replace mutual contrastive loss with mutual distillation (ablation) | CMD · NTU-60 x-sub linear | accuracy: 77.3 -> 79.4 | Improved |
| replace mutual contrastive loss with mutual distillation (ablation) | IMD (DNA) · NTU-60 x-sub KNN | accuracy: 68.0 -> 73.1 | Improved |
| lower teacher temperature to 0.05 with student 0.1 (ablation) | CMD · NTU-60 x-sub linear | accuracy: 77.5 -> 79.8 | Improved |
| raise student temperature from 0.1 to 1.0 (ablation) | CMD · NTU-60 x-sub linear | accuracy: 79.8 -> 76.8 | Hurt |
| vary number of dynamic neighbors from 16 to 128 (ablation) | IMD · NTU-60 x-sub KNN | accuracy: 72.8 -> 73.1 | No clear effect |
Closest papers by meaning
- CMD: Self-Supervised 3D Action Representation Learning with Cross-Modal Mutual Distillation (ECCV 2022 · similarity 0.95)
- Toward Stronger 3D Human Action Representation Learning Via Efficient Spatiotemporal Decoupling (TPAMI 2026 · similarity 0.88)
- Learning an Augmented RGB Representation With Cross-Modal Knowledge Distillation for Action Detection (ICCV 2021 · similarity 0.88)
- Multi-modal Relation Distillation for Unified 3D Representation Learning (ECCV 2024 · similarity 0.87)
- XKD: Cross-Modal Knowledge Distillation with Domain Alignment for Video Representation Learning (AAAI 2024 · similarity 0.87)
- Self-Supervised Learning by Cross-Modal Audio-Video Clustering (NeurIPS 2020 · 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.
