CCGR: Cross-Covariate Gait Recognition

Cross-covariate gait recognition aims to identify people under large appearance and acquisition changes, including clothing, carrying conditions, body-shape variations, viewpoints, and other practical covariates.

This project brings together the CCGR benchmark, CCGR-Mini, and CCGR-3D for cross-covariate gait recognition.

• 2026: 3D Gallery for Cross-Covariate Gait Recognition was published in Knowledge-Based Systems.
• 2024: Cross-Covariate Gait Recognition: A Benchmark was accepted by AAAI 2024.

The released data are provided for non-commercial research use. Please follow the dataset license and cite the corresponding papers when using CCGR, CCGR-Mini, or CCGR-3D.

CCGR contains 970 subjects, about 1.6 million sequences, 33 views, and 53 covariates. It provides multiple gait modalities, including RGB, silhouette, parsing, and pose.

CCGR-Mini is a lightweight subset of CCGR. It keeps 970 subjects, 47,884 sequences, 53 covariates, and 33 views, while retaining one randomly selected view for each covariate of each subject.

CCGR-3D extends CCGR for evaluating cross-covariate gait recognition with 3D data. It provides multiple processed modalities, including silhouette, pose, parsing, SMPL, and skeleton. When using CCGR-3D, please cite both the 2026 KBS paper and the original CCGR benchmark paper.

Silhouette input (%)

Parsing input (%)

Silhouette input (%)

Parsing input (%)

If this project is useful for your research, please cite the following papers.

@article{zou2026_3dgallery,
  title   = {3D Gallery for Cross-Covariate Gait Recognition},
  author  = {Zou, Shinan and Long, Chengyu and Guo, Fan and Tang, Jin},
  journal = {Knowledge-Based Systems},
  volume  = {339},
  pages   = {115576},
  year    = {2026},
  doi     = {10.1016/j.knosys.2026.115576}
}

@article{zou2024_ccgr,
  title   = {Cross-Covariate Gait Recognition: A Benchmark},
  author  = {Zou, Shinan and Fan, Chao and Xiong, Jianbo and Shen, Chuanfu and Yu, Shiqi and Tang, Jin},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  volume  = {38},
  number  = {7},
  pages   = {7855--7863},
  year    = {2024},
  doi     = {10.1609/aaai.v38i7.28621}
}

The dataset is released for non-commercial research use under CC BY-NC-ND unless otherwise specified. Raw RGB data requires a signed usage agreement.

• In the AAAI 2024 version of the CCGR paper, the positions of R5^easy and R5^hard in Table 2 were reversed. The results shown above use the corrected order.
• In Table 3 of the AAAI 2024 version, the batch size of GaitBase and DeepGaitV2 was recorded incorrectly. The corrected results are shown above. These corrections do not affect the conclusions of the paper.

For questions about the dataset or paper, please contact [email protected].