Skip to content

[AAAI 2024] The Official implementation for 'SCD-Net: Spatiotemporal Clues Disentanglement Network for Self-supervised Skeleton-based Action Recognition'.

Notifications You must be signed in to change notification settings

cong-wu/SCD-Net

Folders and files

NameName
Last commit message
Last commit date

Latest commit

0af1b19 · Apr 20, 2024

History

9 Commits
Apr 20, 2024
Mar 21, 2024
Mar 21, 2024
Mar 21, 2024
Mar 21, 2024
Mar 21, 2024
Mar 21, 2024
Mar 21, 2024
Apr 20, 2024
Mar 21, 2024
Mar 21, 2024
Mar 21, 2024
Mar 21, 2024
Mar 21, 2024

Repository files navigation

SCD-Net [AAAI2024]

The Official implementation for 'SCD-Net: Spatiotemporal Clues Disentanglement Network for Self-supervised Skeleton-based Action Recognition' (AAAI 2024). AAAI arXiv

Prerequisite

  • Pytorch

  • We provided requirement file to install all packages, just by running

pip install -r requirements.txt

Data

Generate the data

Download the raw data

Preprocess

  • Preprocess data with python ntu_gendata.py.

Training&Testing

Training

  • To train on NTU-RGB+D 60 under Cross-Subject evaluation, you can run

    python ./pretraining.py --lr 0.01 --batch-size 64 --encoder-t 0.2 --encoder-k 8192 --checkpoint-path ./checkpoints/pretrain/ --schedule 351 --epochs 451 --pre-dataset ntu60 --protocol cross_subject --skeleton-representation joint

Testing

  • For action recognition on NTU-RGB+D 60 under Cross-Subject evaluation, you can run

    python ./action_classification.py --lr 2 --batch-size 1024 --pretrained ./checkpoints/pretrain/checkpoint.pth.tar --finetune-dataset ntu60 --protocol cross_subject --finetune_skeleton_representation joint

  • For action retrieval on NTU-RGB+D 60 under Cross-Subject evaluation, you can run

    python ./action_retrieval.py --knn-neighbours 1 --pretrained ./checkpoints/pretrain/checkpoint.pth.tar --finetune-dataset ntu60 --protocol cross_subject --finetune-skeleton-representation joint

Log files

We also provided some the testing logs in ./log

About

[AAAI 2024] The Official implementation for 'SCD-Net: Spatiotemporal Clues Disentanglement Network for Self-supervised Skeleton-based Action Recognition'.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages