Unofficial Tensorflow implementation of the AAAI'18 paper "Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition"
An unofficial Tensorflow implementation of the paper "Spatial Temporal Graph Convolutional Networks for Skeleton-Based Action Recognition" in AAAI 2018.
Model weights for ST-GCN trained on xview and xsub joint data Dropbox
Most of the interesting stuff can be found in:
model/stgcn.py
: model definition of ST-GCNdata_gen/
: how raw datasets are processed into numpy tensorsgraphs/ntu_rgb_d.py
: graph definitionmain.py
: general training/eval processes; etc.The NTU RGB+D dataset can be downloaded from here. We'll only need the Skeleton data (~ 5.8G).
After downloading, unzip it and put the folder nturgb+d_skeletons
to ./data/nturgbd_raw/
.
Generate the joint dataset first:
cd data_gen
python3 gen_joint_data.py
Specify the data location if the raw skeletons data are placed somewhere else. The default looks at ./data/nturgbd_raw/
.
python3 gen_tfrecord_data.py
To start training the network with the joint data, use the following command:
python3 main.py --train-data-path data/ntu/<dataset folder> --test-data-path data/ntu/<dataset folder>
Here
Note: At the moment, only nturgbd-cross-subject
is supported.
Please cite the following paper if you use this repository in your reseach
@inproceedings{yan2018spatial,
title={Spatial temporal graph convolutional networks for skeleton-based action recognition},
author={Yan, Sijie and Xiong, Yuanjun and Lin, Dahua},
booktitle={Thirty-Second AAAI Conference on Artificial Intelligence},
year={2018}
}