Adaptive Spatiotemporal Graph Transformer Network for Action Quality Assessment

This is the code for "Adaptive Spatiotemporal Graph Transformer Network for Action Quality Assessment".

Framework

Key concept of the proposed ASGTN: the method is to capture the intricate local interactions within an individual clip and across clips in a video, as well as the global contextual semantics information of the entire video. To achieve this, we propose an adaptive spatiotemporal graph specifically designed to capture the subtle interactions. In addition, a transformer encoder is integrated to enhance the long-range dependencies, enriching the video feature representation.

Datasets

• The extracted VST features and label files of Rhythmic Gymnastics and Fis-V datasets can be download from the GDLT repository.

• The original videos of Rhythmic Gymnastics can be downloaded from the ACTION-NET repository.

• The original videos of Fis-V can be downloaded from the MS_LSTM repository.

Running

• Training

python main.py 

• Testing

python main.py --ckpt {pkl file here} --test

The hyper-parameters can be changed in main.py. The path of dataset can be changed in options.py.

Citation

Please cite this work if you find it useful:

@ARTICLE{10884538,
  author={Liu, Jiang and Wang, Huasheng and Zhou, Wei and Stawarz, Katarzyna and Corcoran, Padraig and Chen, Ying and Liu, Hantao},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={Adaptive Spatiotemporal Graph Transformer Network for Action Quality Assessment}, 
  year={2025},
  volume={},
  number={},
  pages={1-1},
  keywords={Transformers;Spatiotemporal phenomena;Feature extraction;Quality assessment;Adaptive systems;Semantics;Long short term memory;Decoding;Data mining;Circuits and systems;Action quality assessment;Graph;transformer;deep learning;neural network},
  doi={10.1109/TCSVT.2025.3541456}}

Acknowledgement

Our code is based on GDLT. Thanks for their great work!

Contact

If you have any questions, please feel free to contact me: [email protected]