TCA-GCN

We are constantly improving the program

Prerequisites

• Python = 3.8.8
• PyTorch = 1.10.0
• Run pip install -e torchlight

NTU RGB+D 60 and 120 and NW-UCLA

1.Download the raw data from the website and place it in the appropriate directory of the './data' file 2.Generate NTU RGB+D 60 and NTU RGB+D 120 dataset: python get_raw_skes_data.py, python get_raw_denoised_data.py, python seq_transformation.py 3. Place the processed data file into the data_path parameter inside the './config'

Training & Testing

Training

Example: training TCAGCN on NTU RGB+D 120 cross subject, the training setup parameters for the other datasets are set under the './config' file python main.py --config config/nturgbd120-cross-subject/default.yaml

Testing

• To test the trained models saved in <work_dir>: python main.py --config <work_dir>/config.yaml --work-dir <work_dir> --weights <work_dir>/.pt

• To ensemble the results of different streams python zhenghe.py

Pretrained Models

We provide individual stream weighting files for the relevant dataset

Acknowledgements

This repo is manly based on CTR-GCN. Many thanks to the authors for their work and to the original authors involved for giving me an opportunity to learn from them.

Citation

Please cite the paper if you find it useful:

@misc{https://doi.org/10.48550/arxiv.2205.15936, doi = {10.48550/ARXIV.2205.15936},

url = {https://arxiv.org/abs/2205.15936},

author = {Wang, Shengqin and Zhang, Yongji and Wei, Fenglin and Wang, Kai and Zhao, Minghao and Jiang, Yu},

keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},

title = {Skeleton-based Action Recognition via Temporal-Channel Aggregation},

publisher = {arXiv},

year = {2022},

copyright = {Creative Commons Attribution Non Commercial No Derivatives 4.0 International} }