Cross-domain Adaptation For Animal Pose Estimation

Jinkun Cao Hongyang Tang Hao-Shu Fang Xiaoyong Shen Cewu Lu Yu-Wing Tai

Updates

• July 22, 2021: We updated the keypoint annotations provided below:

1. We fixed many previous false annotations (in total 260 images), we apologize for our bad annotation quality before.

2. We also provide the images from PASCAL VOC dataset for users' convenience, now all images with keypoint annotations could be downloaded in a file.

3. We aligned the keypoint annotation files with COCO format. Please check the Github repo for more details.

Download

Images in the dataset are collected from different sources, including Internet or other datasets. For users' convenience, we provide a packaged version of dataset image and annotations. For data source and related rights, please check the linked files and acknowledgement.

Part I. Keypoint-labeled animal data (4,000+ images, five categories): [Google Drive]

Part II. Only-bounding-box-labeled animal data (seven categories): [Google Drive] [OneDrive]

Acknowledgement

A collection of keypoint annotations of PASCAL 2011 provided by UC, Berkeley, is the start point of our dataset, on which we provide more annotations and images. Some images are extracted from the Animals-10 dataset. The annotator team of Tencent offered help to build this dataset by annotating most data in it. Other annotation is done by some independent annotators, including our relatives and friends. Many thanks to all of them.

Please follow the citation format. To send any discussion/question or to do contribution, please contact Jinkun Cao (jinkuncao@gmail.com) or Hongyang Tang (thutanghy@gmail.com).

Citation