Human pose estimation based on Latent Tree structure === Datasets === We use the Leeds Sport Dataset [1] to evaluate our algorithm. The training images can be downloaded from http://www.comp.leeds.ac.uk/mat4saj/lsp.html We also use the negative image set in INRIA Person Dataset as negative samples. Please refer to the original datasets and documents on http://pascal.inrialpes.fr/data/human/ === Tools === We adopted Myung Jin Choi's toolbox [2] to generate the latent tree model. The relevent functions are included within the package. The original implementations can be found at http://people.csail.mit.edu/myungjin/latentTree.html We also use liblinear libray to train our visual categories. The tool needs to be downloaded and installed as instructed from http://www.csie.ntu.edu.tw/~cjlin/liblinear/ Furthor more, we build our method based on Yang Yi and Deva Ramanan's work in [3], please refer to the original COPYING_mop and README_mop files for details. === Usage === Please download the datasets and place the image folders in 'data/' directory, name them as follows: data/LEEDSP data/INRIA Compile mex files using the matlab script compile.m, then run demo.m to inspect the algorithm. A pre-trained model is provided in the exprst/ directory. === Acknowledgements === We thanks the authors of the toolboxes and the human pose benchmarks. === Citation === This project is linked to our publication in CVPR 2013, please add the following reference if you used the code: @inproceedings { cvpr2013pose, author = "Fang Wang and Yi Li", title = "Beyond Physical Connections: Tree Models in Human Pose Estimation", booktitle = "IEEE Conference on Computer Vision and Pattern Recognition (CVPR)", address = "Portland, OR, USA " year = "2013", } === Reference === [1] S. Johnson and M. Everingham. Clustered pose and nonlinear appearance models for human pose estimation. BMVC 2010. [2] M. J. Choi, V. Tan, A. Anandkumar, and A. S. Willsky. Learning latent tree graphical models. Journal of Machine Learning 2011. [3] Y. Yang, D. Ramanan. Articulated Pose Estimation using Flexible Mixtures of Parts. CVPR 2011.