We have re-formatted the data for using HUMANISE easier. The re-formatted data can be downloaded from original download link. Please refer to this link if you are the first time to use HUMANISE.
Currently, you just need to download the the motions.zip
and annotation.csv
files. The motions.zip
contains the motion sequence data, and the annotation.csv
contains the annotations for the motion data.
Unzip the motions.zip
file, and you will get the following folder structure:
-| motions
---| 000000.pkl
---| ...
Each *.pkl
store a tuple, in which the first item is numpy array with <L, 159>
shape size storing (trans, orient, body_pose, hand_pose) parameters and the second item is also a numpy array with <10, >
shape size storing the betas parameters.
The annotation.csv
file contains the annotations for the motion data. The content of this file is as follows:
motion_id,scene_id,scene_trans_x,scene_trans_y,scene_trans_z,object_id,object_label,object_semantic_label,action,text
000000,scene0000_00,-4.11121502,-4.17789347,-0.06122726,37,bed,4,lie,lie on the bed
...
Please refer the visualize_demo.py
for more details.
Run the following script to visualize the motion data.
python visualize_demo.py --data_dir ${HUMANISE}$ --scene_dir ${SCANNET}$ --smplx_dir ${SMPLX}$
Download the align_motion_release
and pure_motion
, and put them into the ${HUMANISE}$
folder. Then, run the following script to re-format the data.
python re-format.py --data_dir ${HUMANISE}$ --save_dir ${REFORMAT}$