AGMS-GCN

This repo is the official implementation for AGMS-GCN: Attention-Guided Multi-Scale Graph Convolutional Networks for Skeletal-Based Action Recognition.

Architecture of AGMS-GCN

Data Preparation

There are 3 datasets to download:

• NTU RGB+D 60 Skeleton
• NTU RGB+D 120 Skeleton

Data Processing

Directory Structure

• Put downloaded data into the following directory structure:

  - Ntu60_Skeleton/          # from `nturgbd_skeletons_s001_to_s017.zip`
  - Ntu120_Skeleton/         # from `nturgbd_skeletons_s018_to_s032.zip`
  
  

Generating Data

• Generate NTU RGB+D 60 or NTU RGB+D 120 dataset:

  python ntu60_gen_joint_data.py
  python ntu120_gen_joint_data.py
  python gen_bone_data.py
  python ntu_gen_motion_data.py
  python merge_joint_bone_data.py
  python merge_joint_joint_motion.py
  python merge_bone_bone_motion.py
  python merge_joint_motion_bone_motion.py
  
  

Training & Testing

Training

• You can modify the training and model settings through the configuration files found in the config/ directory. Each dataset has its own dedicated config file, which can be edited to adjust the parameters as needed.
• Load the config file and train the model.
• Cross-view: Train the model with NTU-RGB+D60 coordinate (joint, bone) data.

  python main.py --config config/ntu60_xview.yaml
  

• Cross-subject: Train the model with NTU-RGB+D60 coordinate (joint, bone) data.

  python main.py --config config/ntu60_xsub.yaml  
  
  

Citation

Please cite the following paper if you use this repository in your research.

@article{kilic2025agms,
  title={AGMS-GCN: Attention-guided multi-scale graph convolutional networks for skeleton-based action recognition},
  author={Kilic, Ugur and Karadag, Ozge Oztimur and Ozyer, Gulsah Tumuklu},
  journal={Knowledge-Based Systems},
  pages={113045},
  year={2025},
  publisher={Elsevier}
}

Contact

For any questions, feel free to contact: [email protected]

Acknowledge

This repo is based on ST-GCN, STA-GCN, and 2s-AGCN, thanks to their excellent work.