Animal pose estimation and tracking (APT) is a fundamental task for detecting and tracking
animal keypoints from a sequence of video frames. Previous animal-related datasets focus
either on animal tracking or single-frame animal pose estimation, and never on both
aspects. The lack of APT datasets hinders the development and evaluation of video-based
animal pose estimation and tracking methods, limiting the applications in real world, eg,
understanding animal behavior in wildlife conservation. To fill this gap, we make the first …

ResNet-101 [2] backbone, HRNet-w48 [6], HRFormer-B [10], and ViTPose [8] with ViT-B [1]
backbone. All models are pre-trained with the human pose estimation data from the MS
COCO [5] dataset. As demonstrated in Figure 1, the methods trained with the APT-36K
dataset successfully predict the keypoints of different animal species, despite challenging
cases like occlusion (the 2nd and 3rd row). They can also deal with animals with irregular
body postures like in the 4th row and different orientations like in the 5th row. It can also be …