We use python 3 for this project.
Make sure FFMPEG
is installed, or install by
sudo apt install ffmpeg
Install the python packages dependencies by
pip install -r requirements.txt
Download the data from Total Capture official site.
Please Note that we have NO permission to redistribute this dataset. Please never ask us for a copy.
Then organize the data in below structure (put all data in ./data/totalcapture
):
./data/totalcapture
├── calibration.cal
├── s1
│ ├── acting1
│ │ ├── acting1_BlenderZXY_YmZ.bvh
│ │ ├── acting1_Xsens_AuxFields.sensors
│ │ ├── gt_skel_gbl_ori.txt
│ │ ├── gt_skel_gbl_pos.txt
│ │ ├── s1_acting1_calib_imu_bone.txt
│ │ ├── s1_acting1_calib_imu_ref.txt
│ │ ├── s1_acting1_Xsens.sensors
│ │ ├── TC_S1_acting1_cam1.mp4
│ │ ├── TC_S1_acting1_cam2.mp4
│ │ ├── TC_S1_acting1_cam3.mp4
│ │ ├── TC_S1_acting1_cam4.mp4
│ │ ├── TC_S1_acting1_cam5.mp4
│ │ ├── TC_S1_acting1_cam6.mp4
│ │ ├── TC_S1_acting1_cam7.mp4
│ │ └── TC_S1_acting1_cam8.mp4
│ ├── acting2
│ │ ├── ...
│ ├── ...
├── s2
...
python gendata/gendb.py
# if cannot import tools
# export PYTHONPATH=".:$PYTHONPATH"; python gendata/gendb.py
Options:
You can change options in gendata/config.yaml
, e.g.
- save_frame: whether extract frames from the videos
- gen_train: generate dataset for training
- gen_test: generate dataset for testing
Finally, you should have a bunch of images and two .pkl
files in the ./data/images
. Move the .pkl
files to ../annot
directory, and you get
./data/
├── images
│ ├── s_01_act_01_subact_01_ca_01
│ │ ├── 000000.jpg
│ │ ├── ...
│ ├── ...
├── annot
│ ├── totalcapture_train.pkl
│ ├── totalcapture_validation.pkl
If you want to archive all images in one zip file which is more efficient to be transferred to server or cloud storage, make sure your pwd
is parent directory of images/
, then run zip -0 -r images.zip images/
.