This repo provides the PyTorch implementation of the work:
Learning from Temporal Spatial Cubism for Cross-Dataset Skeleton-based Action Recognition, Yansong Tang*, Xingyu Liu*, Xumin Yu, Danyang Zhang, Jiwen Lu and Jie Zhou (accepted by ACM TOMM) [Paper]
Our code is based on Python3.5. There are a few dependencies to run the code in the following:
- Python (>=3.5)
- PyTorch (0.4.0)
- torchnet
- Visdom
- Other version info about some Python packages can be found in
requirements.txt
We conduct experiments on NTU ↔ PKU, NTU ↔ kinetics, PKU ↔ kinetics, ORGBD ↔ MSRDA3D, and NTU ↔ SBU.
We show the source code of ST-cubsim in this repo, and the code of our compared methods DANN, JAN, CDAN, TA3N, BSP, and GINs.
We present the 51 paired action categories between PKU-MMD and NTU RGB+D in paired_actions.png, and the 12 paired categories between kinetics and NTU RGB+D in paired_actions_between_nk.png.
The experimental results can be referred to Learning from Temporal Spatial Cubism for Cross-Dataset Skeleton-based Action Recognition.
If you find this repo useful in your research, please consider citing:
@article{tang_2022_sda,
author = {Tang, Yansong and Liu, Xingyu and Yu, Xumin and Zhang, Danyang and Lu, Jiwen and Zhou, Jie},
title = {Learning from Temporal Spatial Cubism for Cross-Dataset Skeleton-Based Action Recognition},
year = {2022},
journal = {ACM Transactions on Multimedia Computing, Communications, and Applications (ACM TOMM)},
url = {https://doi.org/10.1145/3472722},
doi = {10.1145/3472722},
}