Action Recognition on MSR Action3D (test)
97.21AccuracyPoint-MAE+ATA
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Point-MAE+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=36 Frames2026.02 | 97.21 | — | — | — | — | — | |
| Point-MAE+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=32 Frames2026.02 | 96.86 | — | — | — | — | — | |
| Point-BERT+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=36 Frames2026.02 | 96.16 | — | — | — | — | — | |
| STS-Mixer2026.04 | 95.85 | — | — | — | — | — | |
| Point-MAE+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=24 Frames2026.02 | 95.62 | — | — | — | — | — | |
| Point-BERT+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=32 Frames2026.02 | 95.47 | — | — | — | — | — | |
| Point-CSAReference=CVPR2025, Learning Protocol=Adaptation for Pre-Trained Models, Number of Frames=32 Frames2026.02 | 95.42 | — | — | — | — | — | |
| Point-CSAReference=CVPR2025, Learning Protocol=Adaptation for Pre-Trained Models, Number of Frames=36 Frames2026.02 | 95.42 | — | — | — | — | — | |
| Point-BERT+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=24 Frames2026.02 | 95.28 | — | — | — | — | — | |
| PointGPT-S+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=24 Frames2026.02 | 95.28 | — | — | — | — | — | |
| Point-CSAReference=CVPR2025, Learning Protocol=Adaptation for Pre-Trained Models, Number of Frames=24 Frames2026.02 | 95.12 | — | — | — | — | — | |
| PointGPT-S+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=32 Frames2026.02 | 95.12 | — | — | — | — | — | |
| PointGPT-S+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=36 Frames2026.02 | 95.12 | — | — | — | — | — | |
| ST-LSTM (Tree) + Trust GateFeature=Geometric2017.06 | 94.8 | — | — | — | — | — | |
| UST-SSM2026.04 | 94.77 | — | — | — | — | — | |
| C2PReference=CVPR2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=24 Frames2026.02 | 94.76 | — | — | — | — | — | |
| Hierarchical RNNFeature=Geometric2017.06 | 94.5 | — | — | — | — | — | |
| Point-MAE+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=16 Frames2026.02 | 94.42 | — | — | — | — | — | |
| MaST-PreReference=ICCV2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=24 Frames2026.02 | 94.08 | — | — | — | — | — | |
| PointGPT-S+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=16 Frames2026.02 | 94.07 | — | — | — | — | — | |
| X4D-SceneFormerReference=AAAI2024, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 93.9 | — | — | — | — | — | |
| LeaFReference=ICCV2023, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 93.84 | — | — | — | — | — | |
| PST-Transformer2026.04 | 93.73 | — | — | — | — | — | |
| PST-TransformerReference=TPAMI2023, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 93.73 | — | — | — | — | — | |
| Point-CSAReference=CVPR2025, Learning Protocol=Adaptation for Pre-Trained Models, Number of Frames=16 Frames2026.02 | 93.73 | — | — | — | — | — | |
| Mamba4D2026.04 | 93.38 | — | — | — | — | — | |
| Point-BERT+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=16 Frames2026.02 | 93.37 | — | — | — | — | — | |
| Kinet2026.04 | 93.27 | — | — | — | — | — | |
| KinetReference=CVPR2022, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 93.27 | — | — | — | — | — | |
| PointCMPReference=CVPR2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=24 Frames2026.02 | 93.27 | — | — | — | — | — | |
| MAMBA4DReference=CVPR2025, Learning Protocol=Supervised Learning, Number of Frames=36 Frames2026.02 | 93.23 | — | — | — | — | — | |
| MAMBA4DReference=CVPR2025, Learning Protocol=Supervised Learning, Number of Frames=32 Frames2026.02 | 93.1 | — | — | — | — | — | |
| CPRReference=AAAI2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=24 Frames2026.02 | 93.03 | — | — | — | — | — | |
| Lillo et al.Feature=Geometric2017.06 | 93 | — | — | — | — | — | |
| Space Time PoseFeature=Geometric2017.06 | 92.8 | — | — | — | — | — | |
| PSTNet++2026.04 | 92.68 | — | — | — | — | — | |
| MAMBA4DReference=CVPR2025, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 92.68 | — | — | — | — | — | |
| PointCPSCReference=ICCV2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=24 Frames2026.02 | 92.68 | — | — | — | — | — | |
| Point-BERT+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=12 Frames2026.02 | 92.68 | — | — | — | — | — | |
| PointGPT-S+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=12 Frames2026.02 | 92.68 | — | — | — | — | — | |
| X4D-SceneFormerReference=AAAI2024, Learning Protocol=Supervised Learning, Number of Frames=16 Frames2026.02 | 92.56 | — | — | — | — | — | |
| Lie GroupFeature=Geometric2017.06 | 92.5 | — | — | — | — | — | |
| PPTrReference=ECCV2022, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 92.33 | — | — | — | — | — | |
| Point-MAE+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=12 Frames2026.02 | 92.33 | — | — | — | — | — | |
| PointCMPReference=CVPR2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=16 Frames2026.02 | 92.26 | — | — | — | — | — | |
| PointCPSCReference=ICCV2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=16 Frames2026.02 | 92.26 | — | — | — | — | — | |
| 3DinAction2026.04 | 92.23 | — | — | — | — | — | |
| CPRReference=AAAI2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=16 Frames2026.02 | 92.15 | — | — | — | — | — | |
| Point-CSAReference=CVPR2025, Learning Protocol=Adaptation for Pre-Trained Models, Number of Frames=12 Frames2026.02 | 92.04 | — | — | — | — | — | |
| PST-TransformerReference=TPAMI2023, Learning Protocol=Supervised Learning, Number of Frames=16 Frames2026.02 | 91.98 | — | — | — | — | — | |
| Point-BERT+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=8 Frames2026.02 | 91.98 | — | — | — | — | — | |
| KinetReference=CVPR2022, Learning Protocol=Supervised Learning, Number of Frames=16 Frames2026.02 | 91.92 | — | — | — | — | — | |
| C2PReference=CVPR2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=16 Frames2026.02 | 91.89 | — | — | — | — | — | |
| PointCMPReference=CVPR2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=12 Frames2026.02 | 91.58 | — | — | — | — | — | |
| LeaFReference=ICCV2023, Learning Protocol=Supervised Learning, Number of Frames=16 Frames2026.02 | 91.5 | — | — | — | — | — | |
| Point-CSAReference=CVPR2025, Learning Protocol=Adaptation for Pre-Trained Models, Number of Frames=8 Frames2026.02 | 91.41 | — | — | — | — | — | |
| Oriented DisplacementsFeature=Geometric2017.06 | 91.3 | — | — | — | — | — | |
| Point-MAE+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=8 Frames2026.02 | 91.28 | — | — | — | — | — | |
| PointGPT-S+ATALearning Protocol=Adaptation for Pre-Trained Models, Number of Frames=8 Frames2026.02 | 91.28 | — | — | — | — | — | |
| PSTNet2026.04 | 91.2 | — | — | — | — | — | |
| PSTNetReference=ICLR2021, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 91.2 | — | — | — | — | — | |
| CPRReference=AAAI2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=12 Frames2026.02 | 91 | — | — | — | — | — | |
| P4Transformer2026.04 | 90.94 | — | — | — | — | — | |
| P4TransformerReference=CVPR2021, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 90.94 | — | — | — | — | — | |
| PPTrReference=ECCV2022, Learning Protocol=Supervised Learning, Number of Frames=16 Frames2026.02 | 90.31 | — | — | — | — | — | |
| PointCPSCReference=ICCV2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=12 Frames2026.02 | 90.24 | — | — | — | — | — | |
| PSTNetReference=ICLR2021, Learning Protocol=Supervised Learning, Number of Frames=16 Frames2026.02 | 89.9 | — | — | — | — | — | |
| PPTrReference=ECCV2022, Learning Protocol=Supervised Learning, Number of Frames=12 Frames2026.02 | 89.89 | — | — | — | — | — | |
| P4TransformerReference=CVPR2021, Learning Protocol=Supervised Learning, Number of Frames=16 Frames2026.02 | 89.56 | — | — | — | — | — | |
| PointCMPReference=CVPR2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=8 Frames2026.02 | 89.56 | — | — | — | — | — | |
| PointCPSCReference=ICCV2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=8 Frames2026.02 | 88.89 | — | — | — | — | — | |
| KinetReference=CVPR2022, Learning Protocol=Supervised Learning, Number of Frames=12 Frames2026.02 | 88.53 | — | — | — | — | — | |
| MeteorNet2026.04 | 88.5 | — | — | — | — | — | |
| MeteorNetReference=ICCV2019, Learning Protocol=Supervised Learning, Number of Frames=24 Frames2026.02 | 88.5 | — | — | — | — | — | |
| SCs (Informative Joints)Feature=Geometric2017.06 | 88.3 | — | — | — | — | — | |
| MeteorNetReference=ICCV2019, Learning Protocol=Supervised Learning, Number of Frames=16 Frames2026.02 | 88.21 | — | — | — | — | — | |
| PST-TransformerReference=TPAMI2023, Learning Protocol=Supervised Learning, Number of Frames=12 Frames2026.02 | 88.15 | — | — | — | — | — | |
| P4TransformerReference=CVPR2021, Learning Protocol=Supervised Learning, Number of Frames=32 Frames2026.02 | 87.93 | — | — | — | — | — | |
| PSTNetReference=ICLR2021, Learning Protocol=Supervised Learning, Number of Frames=12 Frames2026.02 | 87.88 | — | — | — | — | — | |
| P4TransformerReference=CVPR2021, Learning Protocol=Supervised Learning, Number of Frames=12 Frames2026.02 | 87.54 | — | — | — | — | — | |
| C2PReference=CVPR2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=8 Frames2026.02 | 87.16 | — | — | — | — | — | |
| MeteorNetReference=ICCV2019, Learning Protocol=Supervised Learning, Number of Frames=12 Frames2026.02 | 86.53 | — | — | — | — | — | |
| CPRReference=AAAI2023, Learning Protocol=Self-Supervised Learning (End-to-End Fine-Tuning), Number of Frames=8 Frames2026.02 | 86.53 | — | — | — | — | — | |
| X4D-SceneFormerReference=AAAI2024, Learning Protocol=Supervised Learning, Number of Frames=8 Frames2026.02 | 86.47 | — | — | — | — | — | |
| LeaFReference=ICCV2023, Learning Protocol=Supervised Learning, Number of Frames=8 Frames2026.02 | 84.5 | — | — | — | — | — | |
| PPTrReference=ECCV2022, Learning Protocol=Supervised Learning, Number of Frames=8 Frames2026.02 | 84.02 | — | — | — | — | — | |
| PST-TransformerReference=TPAMI2023, Learning Protocol=Supervised Learning, Number of Frames=8 Frames2026.02 | 83.97 | — | — | — | — | — | |
| KinetReference=CVPR2022, Learning Protocol=Supervised Learning, Number of Frames=8 Frames2026.02 | 83.84 | — | — | — | — | — | |
| Joint Angles SimilaritiesFeature=Geometric2017.06 | 83.5 | — | — | — | — | — | |
| PSTNetReference=ICLR2021, Learning Protocol=Supervised Learning, Number of Frames=8 Frames2026.02 | 83.5 | — | — | — | — | — | |
| P4TransformerReference=CVPR2021, Learning Protocol=Supervised Learning, Number of Frames=8 Frames2026.02 | 83.17 | — | — | — | — | — | |
| P4TransformerReference=CVPR2021, Learning Protocol=Supervised Learning, Number of Frames=36 Frames2026.02 | 82.81 | — | — | — | — | — | |
| MeteorNetReference=ICCV2019, Learning Protocol=Supervised Learning, Number of Frames=8 Frames2026.02 | 81.14 | — | — | — | — | — | |
| Histogram of 3D JointsFeature=Geometric2017.06 | 79 | — | — | — | — | — | |
| C2PParadigm=FFT, Pretraining Data=MSR., Tunable Parameters=100%, Memory=N/A2026.06 | — | 87.16 | — | 91.89 | 94.76 | — | |
| CPRParadigm=FFT, Pretraining Data=MSR., Tunable Parameters=100%, Memory=N/A2026.06 | — | 86.53 | 91 | 92.15 | 93.03 | 90.67 | |
| KinetParadigm=SL, Pretraining Data=N/A, Tunable Parameters=N/A, Memory=N/A2026.06 | — | 83.84 | 88.53 | 91.92 | 93.27 | 89.39 | |
| LeaFParadigm=SL, Pretraining Data=N/A, Tunable Parameters=N/A, Memory=N/A2026.06 | — | 84.5 | — | 91.5 | 93.84 | — | |
| MAMBA4DParadigm=SL, Pretraining Data=N/A, Tunable Parameters=N/A, Memory=N/A2026.06 | — | — | — | — | 92.68 | — | |
| MaST-PreParadigm=FFT, Pretraining Data=MSR., Tunable Parameters=100%, Memory=N/A2026.06 | — | — | — | — | 94.08 | — |