Skeleton-based Action Recognition
If you have any problems, suggestions or improvements, please submit the issue or PR.
(New! 2021) PoseC3D 2D Skeleton Dataset (FineGYM, NTURGB-D, Kinetics, Volleyball) [arxiv, Github]
(2017) SYSU 3D Human-Object Interaction Dataset (SYSU)
(2015) UWA3D Multiview Activity II Dataset (UWA3D) [download]
(2014) Northwestern-UCLA Dataset (N-UCLA) [donwload]
This section only shows some popular or new datasets, other available datasets for 3D action recognition and their statistics can be found in the following Table from the journal paper of NTU RGB+D 120 Dataset (TPAMI).
This section only includes the last five papers since 2018 in arXiv.org. Note that arXiv papers without available codes are not included in the leaderboard of performance.
[MV-IGNET] Learning Multi-View Interactional Skeleton Graph for Action Recognition (TPAMI 2020) [paper][Github]
[P&C FW-AEC] PREDICT & CLUSTER: Unsupervised Skeleton Based Action Recognition (CVPR 2020) [paper]
[CA-GC] Context Aware Graph Convolution for Skeleton-Based Action Recognition (CVPR 2020) [paper]
[Shift-GCN] Skeleton-Based Action Recognition With Shift Graph Convolutional Network (CVPR 2020) [paper][Github]
[DMGNN] Dynamic Multiscale Graph Neural Networks for 3D Skeleton Based Human Motion Prediction (CVPR 2020) [paper]
[SGN] Semantics-Guided Neural Networks for Efficient Skeleton-Based Human Action Recognition (CVPR 2020) [arxiv][Github]
[MS-G3D] Disentangling and Unifying Graph Convolutions for Skeleton-Based Action Recognition (CVPR 2020) [arxiv] [Github]
[Dynamic GCN] Dynamic GCN: Context-enriched Topology Learning for Skeleton-based Action Recognition (ACM-MM 2020)[arxiv]
[GCN-NAS] Learning Graph Convolutional Network for Skeleton-based Human Action Recognition by Neural Searching (AAAI 2020) [arxiv] [Github]
[DecoupleGCN-DropGraph] Decoupling GCN with DropGraph Module for Skeleton-Based Action Recognition (ECCV 2020) [arxiv] [Github]
[PA-ResGCN] Stronger, Faster and More Explainable: A Graph Convolutional Baseline for Skeleton-based Action Recognition (ACM-MM 2020) [arxiv] [Github]
[Poincare-GCN] Mix Dimension in Poincaré Geometry for 3D Skeleton-based Action Recognition (ACM-MM 2020) [arxiv]
[STIGCN] Spatio-Temporal Inception Graph Convolutional for Skeleton-Based Action Recognition (ACM-MM 2020) [arxiv]
[JOLO-GCN] JOLO-GCN: Mining Joint-Centered Light-Weight Information for Skeleton-Based Action Recognition (WACV 2021) [arxiv]
[ST-TR-AGCN] Spatial Temporal Transformer Network for Skeleton-based Action Recognition (Under submission at Computer Vision and Image Understanding (CVIU)) [arxiv] [Github]
[PCRP] Prototypical Contrast and Reverse Prediction: Unsupervised Skeleton Based Action Recognition [arxiv] [Github]
The section is being continually updated. We only show results on large-scale dataset NTU-RGB+D and NTU-RGB+D 120.
Year | Methods | Cross-Subject | Cross-View |
---|---|---|---|
2014 | Lie Group | 50.1 | 52.8 |
2015 | H-RNN | 59.1 | 64.0 |
2016 | Part-aware LSTM | 62.9 | 70.3 |
2016 | Trust Gate ST-LSTM | 69.2 | 77.7 |
2017 | Two-stream RNN | 71.3 | 79.5 |
2017 | STA-LSTM | 73.4 | 81.2 |
2017 | Ensemble TS-LSTM | 74.6 | 81.3 |
2017 | Visualization CNN | 76.0 | 82.6 |
2017 | C-CNN + MTLN | 79.6 | 84.8 |
2017 | Temporal Conv | 74.3 | 83.1 |
2017 | VA-LSTM | 79.4 | 87.6 |
2018 | Beyond Joints | 79.5 | 87.6 |
2018 | ST-GCN | 81.5 | 88.3 |
2018 | DPRL | 83.5 | 89.8 |
2019 | Motif-STGCN | 84.2 | 90.2 |
2018 | HCN | 86.5 | 91.1 |
2018 | SR-TSL | 84.8 | 92.4 |
2018 | MAN | 82.7 | 93.2 |
2019 | RA-GCN | 85.9 | 93.5 |
2019 | DenseIndRNN | 86.7 | 93.7 |
2018 | PB-GCN | 87.5 | 93.2 |
2019 | AS-GCN | 86.8 | 94.2 |
2019 | VA-NN (fusion) | 89.4 | 95.0 |
2019 | AGC-LSTM (Joint&Part) | 89.2 | 95.0 |
2019 | 2s-AGCN | 88.5 | 95.1 |
2020 | SGN | 89.0 | 94.5 |
2020 | GCN-NAS | 89.4 | 95.7 |
2019 | 2s-SDGCN | 89.6 | 95.7 |
2019 | DGNN | 89.9 | 96.1 |
2020 | MV-IGNET | 89.2 | 96.3 |
2020 | 4s Shift-GCN | 90.7 | 96.5 |
2020 | DecoupleGCN-DropGraph | 90.8 | 96.6 |
2020 | PA-ResGCN-B19 | 90.9 | 96.0 |
2020 | MS-G3D | 91.5 | 96.2 |
2021 | EfficientGCN-B4 | 91.7 | 95.7 |
2021 | CTR-GCN | 92.4 | 96.8 |
2022 | PoseC3D | 94.1 | 97.1 |
2022 | PSUMNet | 92.9 | 96.7 |
Most of existing methods have not been tested on this new dataset yet, and some results can be found in the paper of NTU RGB+D 120 Dataset (TPAMI).
Year | Methods | Cross-Subject | Cross-Setup |
---|---|---|---|
2019 | SkeleMotion (Magnitude-Orientation) | 62.9 | 63.0 |
2019 | SkeleMotion + Yang et al | 67.7 | 66.9 |
2019 | TSRJI | 67.9 | 59.7 |
2020 | SGN | 79.2 | 81.5 |
2020 | MV-IGNET | 83.9 | 85.6 |
2020 | 4s Shift-GCN | 85.9 | 87.6 |
2020 | DecoupleGCN-DropGraph | 86.5 | 88.1 |
2020 | MS-G3D | 86.9 | 88.4 |
2022 | PoseC3D | 86.9 | 90.3 |
2020 | PA-ResGCN-B19 | 87.3 | 88.3 |
2021 | EfficientGCN-B4 | 88.3 | 89.1 |
2021 | CTR-GCN | 88.9 | 90.6 |
2022 | PSUMNet | 89.4 | 90.6 |