This repository has a toolbox to download, process, and visualize the Panoptic Studio (Panoptic) data.
May-14-2024: We have built a mirror server at SNU.
You may still use the original CMU server, and if the CMU server doesn't respond, You can use the SNU endpoint by simply adding --snu-endpoint option in getData.sh and getData_kinoptic scripts.
Follow these steps to set up a simple example:
git clone https://github.com/CMU-Perceptual-Computing-Lab/panoptic-toolbox cd panoptic-toolbox
To download a dataset, named "171204_pose1_sample" in this example, run the following script.
./scripts/getData.sh 171204_pose1_sample
This bash script requires curl or wget.
This script will create a folder "./171204_pose1_sample" and download the following files.
171204_pose1_sample/hdVideos/hd_00_XX.mp4 #synchronized HD video files (31 views)
171204_pose1_sample/vgaVideos/KINECTNODE%d/vga_XX_XX.mp4 #synchrponized VGA video files (480 views)
171204_pose1_sample/calibration_171204_pose1_sample.json #calibration files
171204_pose1_sample/hdPose3d_stage1_coco19.tar #3D Body Keypoint Data (coco19 keypoint definition)
171204_pose1_sample/hdFace3d.tar #3D Face Keypoint Data
171204_pose1_sample/hdHand3d.tar #3D Hand Keypoint Data
Note that this sample example currently does not have VGA videos.
You can also download any other seqeunce through this script. Just use the the name of the target sequence: instead of the "171204_pose1panopticHD". r example,
./scripts/getData.sh 171204_pose1
for the full version of 171204_pose1 sequence:. You can also specify the number of videospanopticHDnt to donwload.
./scripts/getData.sh (sequenceName) (VGA_Video_Number) (HD_Video_Number)
For example, the following command will download 240 vga videos and 10 videos.
./scripts/getData.sh 171204_pose1_sample 240 10
Note that we have sorted the VGA camera order so that you download uniformly distributed view.
You can find the list of currently available sequences in the following link:
List of released sequences (ver1.2)
Downloading all of them (including videos) may take a long time, but downloading 3D keypoint files (body+face+hand upon their availability) should be "relatively" quick.
You can use the following script to download currently available sequences (ver 1.2):
./scripts/getDB_panopticHD_ver1_2.sh
The default setting is not downloading any videos. Feel free to change the "vgaVideoNum" and "hdVideoNum" in the script to other numbers if you also want to download videos.
You can see the example videos and other information of each sequence: in our website: Browsing dataset.
Check the 3D viewer in each sequence: page where you can visualize 3D skeletons in your web browser. For example: http://domedb.perception.cs.cmu.edu/panopticHDpose1.html
This step requires ffmpeg.
./scripts/extractAll.sh 171204_pose1_sample
This will extract images, for example 171204_pose1_sample/hdImgs/00_00/00_00_00000000.jpg, and the corresponding 3D skeleton data, for example 171204_pose1_sample/hdPose3d_stage1_coco19/body3DScene_00000000.json.
extractAll.sh is a simple script that combines the following set of commands (you shouldn't need to run these again):
cd 171204_pose1_sample ../scripts/vgaImgsExtractor.sh # PNG files from VGA video (25 fps)../scripts/hdImgsExtractor.sh # PNG files from HD video (29.97 fps)tar -xf vgaPose3d_stage1.tar # Extract skeletons at VGA frameratetar -xf hdPose3d_stage1.tar # Extract skeletons for HDcd ..
This codes require numpy, matplotlib.
Visualizing 3D keypoints (body, face, hand):
cd python jupyter notebook demo_3Dkeypoints_3dview.ipynb
The result should look like this.
Reprojecting 3D keypoints (body, face, hand) on a selected HD view:
cd python jupyter notebook demo_3Dkeypoints_reprojection_hd.ipynb
The result should look like this.
This codes require numpy, matplotlib.
Visualizing 3D keypoints (body, face, hand):
cd python jupyter notebook demo_3Dkeypoints_3dview.ipynb
The result should look like this.
Reprojecting 3D keypoints (body, face, hand) on a selected HD view:
cd python jupyter notebook demo_3Dkeypoints_reprojection_hd.ipynb
The result should look like this.
This codes require pyopengl.
Visualizing 3D keypoints (body, face, hand):
python glViewer.py
Note: Matlab code is outdated, and does not handle 3D keypoint outputs (coco19 body, face, hand). Please see this code only for reference. We will update this later.
Matlab example (outdated):
>>> cd matlab>>> demo
We reconstruct 3D skeleton of people using the method of Joo et al. 2018.
The output of each frame is written in a json file. For example,
{ "version": 0.7,
"univTime" :53541.542,
"fpsType" :"hd_29_97",
"bodies" :
[
{ "id": 0,
"joints19": [-19.4528, -146.612, 1.46159, 0.724274, -40.4564, -163.091, -0.521563, 0.575897, -14.9749, -91.0176, 4.24329, 0.361725, -19.2473, -146.679, -16.1136, 0.643555, -14.7958, -118.804, -20.6738, 0.619599, -22.611, -93.8793, -17.7834, 0.557953, -12.3267, -91.5465, -6.55368, 0.353241, -12.6556, -47.0963, -4.83599, 0.455566, -10.8069, -8.31645, -4.20936, 0.501312, -20.2358, -147.348, 19.1843, 0.628022, -13.1145, -120.269, 28.0371, 0.63559, -20.1037, -94.3607, 30.0809, 0.625916, -17.623, -90.4888, 15.0403, 0.327759, -17.3973, -46.9311, 15.9659, 0.419586, -13.1719, -7.60601, 13.4749, 0.519653, -38.7164, -166.851, -3.25917, 0.46228, -28.7043, -167.333, -7.15903, 0.523224, -39.0433, -166.677, 2.55916, 0.395965, -30.0718, -167.264, 8.18371, 0.510041]
}
] }Here, each subject has the following values.
id: a unique subject index within a sequence:. Skeletons with the same id across time represent temporally associated moving skeletons (an individual). However, the same person may have multiple ids joints19: 19 3D joint locations, formatted as [x1,y1,z1,c1,x2,y2,z2,c2,...] where each c ispanopticHDjoint confidence score.
The 3D skeletons have the following keypoint order:
0: Neck 1: Nose 2: BodyCenter (center of hips) 3: lShoulder 4: lElbow 5: lWrist, 6: lHip 7: lKnee 8: lAnkle 9: rShoulder 10: rElbow 11: rWrist 12: rHip 13: rKnee 14: rAnkle 15: lEye 16: lEar 17: rEye 18: rEar
Note that this is different from OpenPose output order, although our method is based on it.
Note that we used to use an old format (named mpi15 as described in our outdated document), but we do not this format anymore.
Kinoptic Studio is a subsystem of Panoptic Studio, which is composed of 10 Kinect2 sensors. Please see: README_kinoptic
You can download all sequences included in our 3D PointCloud DB ver.1 using the following script:
./scripts/getDB_ptCloud_ver1.sh
We have released the processed data for the haggling sequence. Please see Social Signal Processing repository.
Panoptic Studio Dataset is freely available for non-commercial and research purpose only.
By using the dataset, you agree to cite at least one of the following papers.
@inproceedings{Joo_2015_ICCV,
author = {Joo, Hanbyul and Liu, Hao and Tan, Lei and Gui, Lin and Nabbe, Bart and Matthews, Iain and Kanade, Takeo and Nobuhara, Shohei and Sheikh, Yaser},
title = {Panoptic Studio: A Massively Multiview System for Social Motion Capture},
booktitle = {ICCV},
year = {2015} }
@inproceedings{Joo_2017_TPAMI,
title={Panoptic Studio: A Massively Multiview System for Social Interaction Capture},
author={Joo, Hanbyul and Simon, Tomas and Li, Xulong and Liu, Hao and Tan, Lei and Gui, Lin and Banerjee, Sean and Godisart, Timothy Scott and Nabbe, Bart and Matthews, Iain and Kanade, Takeo and Nobuhara, Shohei and Sheikh, Yaser},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2017} }
@inproceedings{Simon_2017_CVPR,
title={Hand Keypoint Detection in Single Images using Multiview Bootstrapping},
author={Simon, Tomas and Joo, Hanbyul and Sheikh, Yaser},
journal={CVPR},
year={2017} }
@inproceedings{joo2019ssp,
title={Towards Social Artificial Intelligence: Nonverbal Social Signal Prediction in A Triadic Interaction},
author={Joo, Hanbyul and Simon, Tomas and Cikara, Mina and Sheikh, Yaser},
booktitle={CVPR},
year={2019}
} 
