Demo Code -- ST-NBNN for Skeleton-based Action Recognition

Requirements

• ANN ( Approximate Nearest Neighbor )
• Linear SVM ( Matlab )
• CVX Toolbox ( Matlab )

Steps

1. In NBNN folder, run make to compile nbnn.cpp file.
2. Run nbnn.exe to generate spatial temporal matrix ( stored in NBNN / MHAD folder ) .
3. Run dataTrans.m in . / NBNN / MHAD folder to transfer ST-matrix to mat format.
4. Copy generated mat file data_X.mat to . / ST-NBNN / MHAD.
5. Run demo.m in ST-NBNN folder.

Attention

• You may need to re-complie the ANN, Liblinear and CVX toolboxes depending on what OS you use.
• The MHAD dataset provided is down-sampled by picking one frame of each 20 frames due to the large size. The expected results would be 89.1% for NBNN and 100% for ST-NBNN.

Citation

Please cite the following paper if you use this source code in your research.

@InProceedings{Weng_2017_CVPR,  
author    = {Weng, Junwu and Weng, Chaoqun and Yuan, Junsong},  
title     = {Spatio-Temporal Naive-Bayes Nearest-Neighbor (ST-NBNN) for Skeleton-Based Action Recognition},  
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},  
month     = {July},  
year      = {2017}  
}