umariqb/action4pose_fg2017

Source code for our FG'17 paper: "Pose for Action - Action for Pose"

Action4Pose_FG2017

Source code for our FG'17 paper:

Umar Iqbal, Martin Garbade, Juergen Gall
Action for Pose - Pose for Action
IEEE Conference on Automatic Face and Gesture Recognition (FG) 2017.

For more information visit http://pages.iai.uni-bonn.de/iqbal_umar/action4pose/

Installation

Dependencies

• OpenCV
• Boost-1.55
• Caffe
• CUDA >= 7
• LibConfig++
• GFlags
• GLOG

Installation Instructions

1. Clone repository

   $ git clone https://github.com/iqbalu/Action4Pose_FG2017.git --recursive
   

2. Download models

   $ cd data
   $ ./download_models.sh
   

3. Download and build Caffe. Follow Caffe building instructions.
   You may also need to run the following commands to connect Caffe with the code using CMake.

   $ cd [caffe_dir]
   $ mkdir cmake_build && cd cmake_build
   $ cmake .. -DBUILD_SHARED_LIB=ON
   $ cmake . -DCMAKE_BUILD_TYPE=Release   
   $ make -j 12 && make install          
   

4. Build the code

$ cd demo_PennAction && mkdir build && cd build
$ cmake ..
$ make

5. Running

$ ./bin/demo_PennAction config_file.txt

Parameters in Config file:

• save: set to True if you want to save the poses in a text file. The text file will be stored in cache folder. Set it to False if you only want to visualize the results.
• Other parameters are self explanatory.

Citing

@inproceedings{iqbal2017FG,
	author = {Umar Iqbal and Martin Garbade, Juergen Gall},
	title = {Action for Pose - Pose for Action},
	booktitle = {IEEE Conference on Automatic Face and Gesture Recognition},
	year = {2017},
	url = {https://arxiv.org/pdf/1603.04037.pdf}
}