Requirements

SSRL

This is an official PyTorch implementation of "SSRL: Self-supervised Spatial-temporal Representation Learning for 3D Action recognition"

Download the raw data of NTU RGB+D and PKU-MMD.
For NTU RGB+D dataset, preprocess data with tools/ntu_gendata.py. For PKU-MMD dataset, preprocess data with tools/pku_part1_gendata.py.
Then downsample the data to 50 frames with feeder/preprocess_ntu.py and feeder/preprocess_pku.py.

# Install torchlight
$ cd torchlight
$ python setup.py install
$ cd ..

# Install other python libraries
$ pip install -r requirements.txt

Example for unsupervised pre-training of SSRL. You can change some settings of .yaml files in config folder.

# train on NTU RGB+D xsub joint stream
$ python main.py pretrain_SSRL --config config/pretext_sticlr_xsub_joint.yaml

Example for linear evaluation of SSRL. You can change .yaml files in config folder.

# Linear_eval on NTU RGB+D xsub joint stream
$ python main.py linear_evaluation --config config/ntu60/linear_eval/linear_eval_sticlr_xsub_joint.yaml

The t-SNE visualization of the embeddings after SSRL pre-training on NTU60-xsub.

Pre-trained models on NTU-60 are provided in weights folder. You can download and test them with linear evaluation by changing the config file.

The framework of our code is extended from the following repositories. We sincerely thank the authors for releasing the codes.