MS2L

MS2L: Multi-Task Self-Supervised Learning for Skeleton Based Action Recognition in ACMMM 2020

Lilang Lin, Sijie Song, Wenhan Yang, Jiaying Liu

[Project Page]|[Paper]

Data Prepreprocessing Instructions

Apply the dataset processing as SkeletonContrast.

Training & Testing

Pretrain with Multi-Task Self-Supervised Learning (MS^2L):

python procedure.py with 'train_mode="pretrain"'

Finetune with labeled data:

python procedure.py with 'train_mode="loadweight_linear"'

python procedure.py with 'train_mode="loadweight_finetune"'

Citation

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

@inproceedings{lin2020ms2l,
    title       = {MS2L: Multi-Task Self-Supervised Learning for Skeleton Based Action Recognition},
    author      = {Lin, Lilang and Song, Sijie and Yang, Wenhan and Liu, Jiaying},
    booktitle   = {Proceedings of the 28th ACM International Conference on Multimedia},
    pages       = {2490--2498},
    year        = {2020}
}

Contact

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