healthonrails/mouse_keypoints

YOLO Pose Dataset (ADPT-derived)

This dataset contains images and YOLO-format pose annotations for training Ultralytics YOLO pose models. The original images/labels are derived from the ADPT dataset (see source below). This release provides a YOLO-ready layout plus a simple data.yaml for immediate training and evaluation.

Source / Attribution

Original dataset: ADPT (Tang Guoling et al.) Upstream repository: https://github.com/tangguoling/ADPT

This Hugging Face dataset is a repackaged and converted version intended for convenient use with Ultralytics YOLO. Please refer to the ADPT repository for upstream details, original licensing/terms, and citations.

Dataset Structure

dataset/
├── images/
│   ├── train/
│   ├── val/
│   └── test/        # optional
├── labels/
│   ├── train/
│   ├── val/
│   └── test/        # optional
├── data.yaml
└── README.md

• —images/** contains the image files (.jpg / .png)
• —labels/** contains YOLO pose label files (.txt), one per image

Annotation Format (YOLO Pose)

Each label line corresponds to one instance:

class x_center y_center width height x1 y1 v1 x2 y2 v2 ... xN yN vN

All coordinates are normalized to [0,1].

Visibility flag v:

• —0: not labeled
• —1: labeled but not visible
• —2: labeled and visible

data.yaml (Example)

Update class names and keypoints to match your dataset:

path: .
train: images/train
val: images/val

nc: 1
names: ["object"]

kpt_shape: [K, 3]      # K = number of keypoints
flip_idx: []           # optional, set if left/right keypoints exist

Train with Ultralytics YOLO

yolo pose train model=yolo11n-pose.pt data=data.yaml imgsz=640 epochs=300

Inference:

yolo pose predict model=best.pt source=images/val

Notes

• —This repo focuses on YOLO compatibility (folder layout + label format).
• —If you need the original ADPT annotation format or metadata, use the upstream repo.

License

The license of the original data is defined by ADPT. Please follow the upstream dataset’s license and attribution requirements:

https://github.com/tangguoling/ADPT

Citation

If you use this dataset, please cite the original ADPT work (source dataset) and Annolid (tooling / conversion / workflow), as appropriate.

ADPT (Upstream Dataset / Method) @article{tang2025adpt, title = {Anti-drift pose tracker (ADPT), A transformer-based network for robust animal pose estimation cross-species}, author = {Tang, Guoling and Han, Yaning and Sun, Xing and Zhang, Ruonan and Han, Ming-Hu and Liu, Quanying and Wei, Pengfei}, journal = {eLife}, year = {2025}, month = {03}, doi = {10.7554/eLife.95709} } Annolid (Tooling / Conversion / Pipeline) @misc{yang2024annolid, title = {Annolid: Annotate, Segment, and Track Anything You Need}, author = {Chen Yang and Thomas A. Cleland}, year = {2024}, eprint = {2403.18690}, archivePrefix = {arXiv}, primaryClass = {cs.CV} }

@article{yang2023automated, title = {Automated Behavioral Analysis Using Instance Segmentation}, author = {Yang, Chen and Forest, Jeremy and Einhorn, Matthew and Cleland, Thomas A}, journal = {arXiv preprint arXiv:2312.07723}, year = {2023} }

@misc{yang2020annolid, author = {Chen Yang and Jeremy Forest and Matthew Einhorn and Thomas Cleland}, title = {Annolid: an instance segmentation-based multiple animal tracking and behavior analysis package}, howpublished = {\url{https://github.com/healthonrails/annolid}}, year = {2020} }