Henu-Software/MultiSubjects-Gait

Our paper:

《STGGait: A Graph Transformer Network for Pose-based Gait Recognition》,

has been accepted by: IEEE International Conference on Multimedia & Expo 2025 (ICME 2025)!

We will make the dataset publicly available after the conference.

Dataset Card for MultiSubjects-Gait

MutiSubjects-Gait is built based on MutiSubjects.

MutiSubjects-Gait is the first dataset available for identity recognition based on basketball actions, exploring the application of gait recognition in complex sports scenarios.

We divide it into three subsets according to different actions: MutiSubjects-D, MutiSubjects-P, and MutiSubjects-S.

We use HRNet to extract 2D human keypoint data and align the structure of the dataset with CASIA-B Pose to facilitate experimental comparison.

Dataset Information

MutiSubjects-D:

{
Action: Dribbling

Number of subjects: 424

Training ID: 001-324

Test ID：325-424

Gallery: #01

Probe: #02-03
}

MutiSubjects-P:

{
Action: Layup

Number of subjects: 209

Training ID: 001-159

Test ID：160-209

Gallery: #01

Probe: #02
}

MutiSubjects-S:

{
Action: Shooting

Number of subjects: 659

Training ID: 001-527

Test ID：528-659

Gallery: #01

Probe: #02-03
}

Dataset Curators

Authors of STGGait

• Wansong Qin
• Zhijie Han
• Yaru Li

Citation Information

......