Home

Awesome

ShiftGCN++

The implementation for "Extremely Lightweight Skeleton-Based Action Recognition with ShiftGCN++" (TIP2021). ShiftGCN++ further boosts the efficiency of ShiftGCN, which achieves comparable performance with 6× less FLOPs and 2× practical speedup.

image

image

image

Prerequisite

Compile cuda extensions

cd ./model/Temporal_shift
bash run.sh

Data Preparation

Training & Testing

Multi-stream ensemble

To ensemble the results of 4 streams. Change models name in ensemble.py depending on your experiment setting. Then run python ensemble.py.

Trained models

We release several trained models:

ModelDatasetSettingTop1(%)
./save_models/ntu_ShiftGCN-plus_joint_xview.ptNTU-RGBDX-view94.8
./save_models/ntu_ShiftGCN-plus_bone_xview.ptNTU-RGBDX-view94.7
./save_models/ntu_ShiftGCN-plus_joint_xsub.ptNTU-RGBDX-sub87.9
./save_models/ntu_ShiftGCN-plus_bone_xsub.ptNTU-RGBDX-sub88.3

Citation

If you find this model useful for your research, please use the following BibTeX entry.

@article{cheng2021extremely,
title={Extremely Lightweight Skeleton-Based Action Recognition With ShiftGCN++},
author={Cheng, Ke and Zhang, Yifan and He, Xiangyu and Cheng, Jian and Lu, Hanqing},
journal={IEEE Transactions on Image Processing},
volume={30},
pages={7333--7348},
year={2021},
publisher={IEEE}
}