- ANN ( Approximate Nearest Neighbor )
- Linear SVM ( Matlab )
- CVX Toolbox ( Matlab )
- In NBNN folder, run
make
to compilenbnn.cpp
file. - Run
nbnn.exe
to generate spatial temporal matrix ( stored in NBNN / MHAD folder ) . - Run
dataTrans.m
in . / NBNN / MHAD folder to transfer ST-matrix to mat format. - Copy generated mat file
data_X.mat
to . / ST-NBNN / MHAD. - Run
demo.m
in ST-NBNN folder.
- 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.
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}
}