Research map / Video / Action recognition
Masked Motion Predictors are Strong 3D Action Representation Learners
Yunyao Mao, Jiajun Deng, Wengang Zhou, Yao Fang, 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 |
|---|---|---|---|
| predict motion (temporal difference) of masked joints instead of reconstructing joint coordinates (ablation) | MAMP · NTU-60 X-sub (linear eval) | accuracy: 74.8 -> 84.9 | Improved |
| predict motion instead of joints (ablation) | MAMP · NTU-120 X-sub (linear eval) | accuracy: 72.5 -> 78.6 | Improved |
| feed motion instead of joints as input (ablation) | MAMP · NTU-60 X-sub (linear eval) | accuracy: 84.9 -> 75.9 | Hurt |
| use motion-aware masking instead of random masking (ablation) | MAMP · NTU-60 X-sub (linear eval) | accuracy: 83.7 -> 84.9 | Improved |
| use motion-aware masking instead of random masking (ablation) | MAMP · NTU-120 X-sub (linear eval) | accuracy: 77.3 -> 78.6 | Improved |
| add MAMP pre-training before fine-tuning vanilla transformer | Transformer · NTU-60 X-sub (fine-tuned) | accuracy: 83.1 -> 93.1 | Improved |
| add MAMP pre-training before fine-tuning vanilla transformer | Transformer · NTU-120 X-sub (fine-tuned) | accuracy: 76.8 -> 90.0 | Improved |
| add MAMP pre-training with 1% labeled data | Transformer · NTU-60 X-sub (semi-supervised, 1% labels) | accuracy: 38.8 -> 66.0 | Improved |
| use MAMP instead of SkeletonMAE* for pre-training under linear evaluation | Transformer · NTU-120 X-set (linear eval) | accuracy: 73.5 -> 79.1 | Improved |
| MAMP versus PoseC3D in fully supervised NTU-60 X-sub | MAMP · NTU-60 X-sub (supervised, no ensemble) | accuracy: 93.1 vs 93.7 | Hurt |
Closest papers by meaning
- Rethinking Masked Data Reconstruction Pretraining for Strong 3D Action Representation Learning (AAAI 2025 · similarity 0.91)
- Exploring Adaptive Masked Reconstruction for Self-Supervised Skeleton-Based Action Recognition (CVPR 2026 · similarity 0.90)
- Self-Supervised Action Representation Learning from Partial Spatio-Temporal Skeleton Sequences (AAAI 2023 · similarity 0.89)
- A Unified Masked Autoencoder with Patchified Skeletons for Motion Synthesis (AAAI 2024 · similarity 0.87)
- MMM: Generative Masked Motion Model (CVPR 2024 · similarity 0.87)
- MaskCLR: Attention-Guided Contrastive Learning for Robust Action Representation Learning (CVPR 2024 · 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.
