Home

Awesome

Skeleton-based Action Recognition Papers and Small Notes About Them

I am keeping these notes for my research at Fraunhofer IPA. For each paper, I am planning to give a link, accuracy on the NTU-RGBD dataset and some small notes.

Contribution

Feel free to contribute. No general rule. Just keep the format for each paper as below.

Template:
**Name of the paper**
Link: 
Code:
Accuracy on Cross Subject NTU-RGBD: **XX%**
Notes:
- Bullet point 1
- Bullet point 2

Current Top 2 for NTU-RGBD Cross Subject Split: (Only using Skeleton data, not RGBD)

Top 1Top 2
Accuracy:0.8990.894
Link:LinkLink

Papers:

1. SKELETON-BASED ACTION RECOGNITION WITH CONVOLUTIONAL NEURAL NETWORKS

Link: https://arxiv.org/abs/1704.07595

Code:

Accuracy on Cross Subject NTU-RGBD: 0.832

Notes:

2. Co-occurrence Feature Learning from Skeleton Data for Action Recognition and Detection with Hierarchical Aggregation

Link: https://arxiv.org/abs/1804.06055

Code: https://github.com/huguyuehuhu/HCN-pytorch (Re-implementation PyTorch Accuracy is %1.5 lower than original)

Accuracy on Cross Subject NTU-RGBD: 0.865

Notes:

3. Skeleton-Based Action Recognition with Synchronous Local and Non-local Spatio-temporal Learning and Frequency Attention

Link: https://arxiv.org/abs/1811.04237

Code:

Accuracy on Cross Subject NTU-RGBD: 0.891

Notes:

4. Interpretable 3D Human Action Analysis with Temporal Convolutional Networks

Link: https://arxiv.org/abs/1704.04516

Code: https://github.com/TaeSoo-Kim/TCNActionRecognition

Accuracy on Cross Subject NTU-RGBD: 0.743

Notes:

5. Ensemble One-Dimensional Convolution Neural Networks for Skeleton-Based Action Recognition

Link: https://arxiv.org/abs/1801.02475 (but no pdf!)

Code: https://github.com/Qingyang-Xu/Ensem-NN

Accuracy on Cross Subject NTU-RGBD: 0.851

Notes:

6. Hard Sample Mining and Learning for Skeleton-Based Human Action Recognition and Identification

Link: https://ieeexplore.ieee.org/abstract/document/8588326

Code:

Accuracy on Cross Subject NTU-RGBD: 0.866

Notes:

7. View Adaptive Neural Networks for High-Performance Skeleton-based Human Action Recognition

Link: https://arxiv.org/abs/1804.07453

Code: https://github.com/microsoft/View-Adaptive-Neural-Networks-for-Skeleton-based-Human-Action-Recognition

Accuracy on Cross Subject NTU-RGBD: 0.894

Notes:

8. Actional-Structural Graph Convolutional Networks forSkeleton-based Action Recognition

Link: https://arxiv.org/pdf/1904.12659.pdf

Code: https://github.com/limaosen0/AS-GCN

Accuracy on Cross Subject NTU-RGBD: 0.861

Notes:

9. Skeleton-Based Action Recognition with Directed Graph Neural Networks

Link: http://openaccess.thecvf.com/content_CVPR_2019/papers/Shi_Skeleton-Based_Action_Recognition_With_Directed_Graph_Neural_Networks_CVPR_2019_paper.pdf

Code:

Accuracy on Cross Subject NTU-RGBD: 0.899

Notes:

10. Make Skeleton-based Action Recognition ModelSmaller, Faster and Better

Link: https://arxiv.org/abs/1907.09658

Code: https://github.com/fandulu/DD-Net

Accuracy on Cross Subject NTU-RGBD: Not tested on NTU-RGBD

Notes:

11. Deep Independently Recurrent Neural Network (IndRNN)

Link: https://arxiv.org/abs/1910.06251

Code: https://github.com/Sunnydreamrain/IndRNN_pytorch

Accuracy on Cross Subject NTU-RGBD: 0.867

Notes:

12. Two-Stream Adaptive Graph Convolutional Networks for Skeleton-Based Action Recognition

Link: http://openaccess.thecvf.com/content_CVPR_2019/papers/Shi_Two-Stream_Adaptive_Graph_Convolutional_Networks_for_Skeleton-Based_Action_Recognition_CVPR_2019_paper.pdf

Code: https://github.com/lshiwjx/2s-AGCN

Accuracy on Cross Subject NTU-RGBD: 0.885

Notes:

Other Github Repos for Skeleton-based Action Recognition Papers

Websites for Skeleton-based Action Recognition Papers

Acknowledgement

This work(Github REPO) has received funding from the European Unions Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 721619 for the SOCRATES project.