metadata
extra_gated_prompt: >-
Please provide the information below to request access. Requests are reviewed
manually.
extra_gated_fields:
Full name: text
Work email: text
Affiliation or organization: text
What are you planning to use it for?: text
I agree to use this dataset only for research and not redistribute it:
type: checkbox
BarbieGait: An Identity-Consistent Synthetic Human Dataset with Versatile Cloth-Changing for Gait Recognition
CVPR 2026
Qingyuan Cai · Saihui Hou · Xuecai Hu · Yongzhen Huang*
School of Artificial Intelligence, Beijing Normal University · AMAP, Alibaba Group · WATRIX.AI
Dataset Access
The BarbieGait dataset is hosted on Hugging Face. Please fill out the access request manually. We will handle your requests within a week. In case you encounter any issues, please feel free to reach out to us via caiqingyuan@mail.bnu.edu.cn.
After obtaining the data, follow the data preparation guide to preprocess it.
Training on BarbieGait
Prepare the required P2 data view before training. The data preparation guide covers silhouette P2 links and pose-to-heatmap conversion.
python datasets/create_symlnk.py --modality sil
Predicted Silhouette
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch \
--master_port 13359 --nproc_per_node=8 opengait/main.py \
--cfgs ./configs/gaitclif/GaitCLIF_BarbieGait_predsil_10layer_p3d_261p.yaml \
--phase train --log_to_file
Predicted Pose Heatmaps
CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch \
--master_port 13359 --nproc_per_node=8 opengait/main.py \
--cfgs ./configs/gaitclif/GaitCLIF_BarbieGait_predpose_10layer_261p.yaml \
--phase train --log_to_file
✅ TODO
- Release the paper link
- Release the BarbieGait predicted silhouette and 2D pose
- Release the GaitCLIF codebase
- Improve documentation and usage examples
- Release the BarbieGait rendered ground truth silhouette and rendered RGB data