aihgf / deepcut

Multi Person Pose Estimation

Deep(er)Cut: Multi Person Pose Estimation

Here you can find the code that implements DeepCut and DeeperCut papers.

Installation instructions

Only Linux 64bit is supported.

Prerequisites:

• HDF5 1.8
• CMake
• C++ 11
• CUDA >=7.5
• Caffe building instructions
• Gurobi optimizer 6.0.x

$ git clone https://github.com/eldar/deepcut --recursive

# build Caffe and its Matlab interface, for example, after properly configuring Makefile.config:
$ cd pose/external/caffe
$ make -j 4 all matcaffe

# build liblinear, specify the path to the Matlab installation
$ cd ../liblinear-1.94/matlab
$ CC=gcc CXX=g++ MATLABDIR=/usr/lib/matlab-8.6/ make

$ cd ../../solver
$ cmake . -DGUROBI_ROOT_DIR=/some/path/gurobi603/linux64 -DGUROBI_VERSION=60
$ make solver-callback

# Download models
$ cd <root_dir>/data
$ ./download_models.sh

# Obtain Gurobi license from http://www.gurobi.com/downloads/licenses/license-center
# and place the license file license.lic in data/gurobi or modify parameter 
# p.gurobi_license_file in lib/pose/exp_params.m to point to the license file location

$ cd <root_dir>
$ ./start_matlab.sh

In Matlab:

% Demo multi person pose estimation
demo_multiperson()

CNN-based part detectors

If you are interested in trying out our part detectors that produce dense confidence maps, check out the respective project page.

Citation

Please cite Deep(er)Cut in your publications if it helps your research:

@article{insafutdinov2016deepercut,
    author = {Eldar Insafutdinov and Leonid Pishchulin and Bjoern Andres and Mykhaylo Andriluka and Bernt Schiele},
    url = {http://arxiv.org/abs/1605.03170}
    title = {DeeperCut: A Deeper, Stronger, and Faster Multi-Person Pose Estimation Model},
    year = {2016}
}

@inproceedings{pishchulin16cvpr,
    title = {DeepCut: Joint Subset Partition and Labeling for Multi Person Pose Estimation},
    booktitle = {CVPR'16},
    url = {},
    author = {Leonid Pishchulin and Eldar Insafutdinov and Siyu Tang and Bjoern Andres and Mykhaylo Andriluka and Peter Gehler and Bernt Schiele}
}