####################
MPI-INF-3DHP Dataset
####################

1. 
download the prepare dataset zip package, MPI-INF-3DHP Dataset zip from: 
http://gvv.mpi-inf.mpg.de/3dhp-dataset/

2.
unzip, then read the MPI-INF-3DHP README.txt file, attention:
Use the script get_dataset.sh to download the training set and get_testset.sh for the test set. 
You would need to read and review the configuration under conf.ig before you can proceed with downloading the dataset !!!

Make sure you have approx 25GB space in this path to download the complete training set.
The test set needs another 7GB and can be downloaded with get_testset.sh

change these configs in conf.ig for running shell script
subjects=(1 2 3 4 5 6 7 8)
ready_to_download=1

3. 
too much times, use shell script with this command: 
nohup sh get_dataset.sh &

4. select the directory that contains README.txt, S1..S8, etc

The image frames of the dataset come in the form of video sequences!!!
using ffmpeg (ffmpeg -i "<some_folder>/video_X.avi" -qscale:v 1 "<some_folder>/img_X_%%06d.jpg") to ensure valid correspondence between the annotations and the frames.

using the script get_dataset_img.sh to get train images from video sequences.

ffmpeg:
ffmpeg -i *.avi -vf "select=between(n\,84\,208)*not(mod(n\,25))" -vsync 0 ./images/image_%06d.jpg
  -vf select filter, between(n,*) means split from 84 frame to 208 frame
                     not(mode(n\, K)) means output 1 image from every 25 frames 
ffmpeg -i *.avi -r 1 -vf fps=fps=1 ./images/image_%06d.jpg
    -vf fps=fps=1(-r 1) means the rate of screenshots is 1 frame per second