This repo maintains the animal pose dataset proposed in paper Cross-Domain Adaptation for Animal Pose Estimation on ICCV'2019. The project page is here.
The dataset splits to two subsets:
- Part I: includes animals (of five categories: cow, sheep, horse, cat, dog) with both bounding boxes and keypoints annotated. [Google Drive]
- Part II: includes animals of other seven species, provided for unsupervised domain adptation task. [Google Drive]
Please contact us if you have trouble downloading the files from Google Drive.
For the Part I subset, the annotations follow the format convention of COCO. For each instance, it has bounding box annotation in the format [xmin, ymin, xmax, ymax] and the keypoint annotation in the format [x, y, visible]. If the visible flag of a keypoint is 1, it is annotated and shouble not be occluded.
For the Part II subset, the annotation is also included in a json file but only bounding boxes coordinates [xmin, ymin, xmax, ymax] are provided.
If the dataset helps you, please cite our work:
@inproceedings{cao2019cross, title={Cross-domain adaptation for animal pose estimation}, author={Cao, Jinkun and Tang, Hongyang and Fang, Hao-Shu and Shen, Xiaoyong and Lu, Cewu and Tai, Yu-Wing}, booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision}, pages={9498--9507}, year={2019} }