HKBU-Mocap dataset

Introduction

HKBU-Mocap consists of 5 video sequences with multi-view multi-person motions. We now release one sequence of HKBU-Mocap presented in our paper, called COOP-6 (short for 6th cooperation). COOP-6 supports 900 frames with 5 camera observations under 3-person performances. It is manually annotated and the annotation format is BODY25 given by OpenPose (see the corresponding skeleton definition here). Currently, the data is uploaded to Google Drive and Baidu Drive and the project page is a GitPage (we are trying to make it more professional). For more details, please refer to our supplementary material. Since the manual annotation is hard work, some of them are still not perfect. It affects by the inconsistent camera parameters, the artifacts given by humans and also the unstable capturing settings. However, we will continue to maintain the dataset and provide more idealistic data.

Browse

Twist
Twist
Three actors interact and twist with 5 observations. Several challenging poses appear. #multi-person #multi-view

Download: Google Drive Baidu Drive

Specs:

    Resolution: 3840 x 2160
    Frame length: 1200
    Annotation length: 900
    Video Encode CRF: 30

File Structure:

    Annotation: coop_6_3D.npy
    Video Sequences: video.zip
    Frame RGB images: image.zip
    Camera parameters: camera_params.npy

Citation

1
2
3
4
5
6
@inproceedings{jiang2022dmae,
  title={A Dual-Masked Auto-Encoder for Robust Motion Capture with Spatial-Temporal Skeletal Token Completion},
  author={Jiang, Junkun and Chen, Jie and Guo, Yike},
  booktitle={Proceedings of the 30th ACM international conference on Multimedia},
  year={2022}
}