Structured Feature Learning for Pose Estimation

This is the code for our work Structured Feature Learning for Pose estimation

Training

Make caffe: We write our own layer for loss, channel dropout and mix interpolation, if you are not going to use these functions, you can use your own caffe.

make matcaffe

Get LMDB: Run "Data_prepare.m" in matlab to generate LMDB requires Train the caffe model: Run "Baseline.sh. You may need the pre-train fully convolutional VGG-16 model.

./Baseline.sh

Test: Select the best model for testing, and run "TestModel.m" to see the results.

Released models

We provide a model we trained on LSP dataset (itration = 3250). If you are going to test this model, please download it and put it in the location specified in code, and set the variable "test_our_provided_model" to true.

Cite

If you use this code, please cite our work

@inproceedings{chu2016structure, 
title={Structured Feature Learning for Pose Estimation}, 
author={Chu, Xiao and Ouyang, Wanli and Li,Hongsheng and Wang, Xiaogang}, 
booktitle={CVPR}, year={2016} 
}

Our project is written based on Xianjie Chen's NIPS2014

About

This is the code for our work "Structured Feature Learning for Pose estimation"

Languages

Language:Jupyter Notebook 50.7%Language:C++ 37.8%Language:Python 4.6%Language:Cuda 2.5%Language:MATLAB 1.5%Language:CMake 1.5%Language:Protocol Buffer 0.7%Language:Makefile 0.3%Language:Shell 0.3%Language:CSS 0.1%Language:HTML 0.1%Language:M 0.0%