HKBU-Mocap dataset

Introduction

HKBU-Mocap consists of 5 video sequences featuring multi-view, multi-person motions. We have now released all of these sequences, referred to as COOP-6 (short for 6th cooperation), as presented in our paper. COOP-6 provides complete pose annotations for all persons in every frame—there are 2,700 frames in total (originally recorded at 30 FPS and downsampled to 900 frames for easier storage)—captured from 5 cameras during a 3-person activity. All individuals appearing in each frame are manually annotated, following the BODY25 format as defined by OpenPose (see the corresponding skeleton definition here).

Currently, the data is available for download via Google Drive and Baidu Drive. The project page is hosted on a GitPage (we are trying to make it more professional). More details can be found in our supplementary material. Please note that, due to the challenging nature of manual annotation, some imperfections remain. These are mainly caused by inconsistent camera parameters, annotation artifacts, and instabilities in the capture setup. Nevertheless, we are committed to maintaining the dataset and plan to provide improved and more comprehensive data in the future.

Browse

Union

Three actors interact with each other and form a circle together. #multi-person #multi-view

Spread

Three actors shake hands with each other and then move away from one another. #multi-person #multi-view

Across

One actor walks across the two other actors, who are holding hands to form a line. #multi-person #multi-view

Rotate

Three actors pair up to hold each other's hands aloft, with one member of each pair rotating. #multi-person #multi-view

Through

Two actors clasp hands to form a bridge, while another actor passes underneath. #multi-person #multi-view

❮ ❯

Download links: Google Drive | Baidu Drive

Specs:

Resolution: 3840 x 2160

Video length: 2 minutes

Annotated frame number: 2700

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

@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}
}

---