Body-Part Guided Animal Pose Estimation

Jiyong Rao, Tianyang Xu, Xiaoning Song, Zhenhua Feng, Xiaojun Wu

Abstract: In this paper, we present our solution to ICME Grand Challenge in animal pose estimation track in the 2024 Multi-Modal Video Reasoning and Analyzing Competition (MMVRAC 2024). We introduce a novel body-part guided framework for cross-species animal pose estimation task, which consists of two coupled components. The first component is to extract keypoint features and generate body part visual prompts. And the second is a Keypoint-Interactive Transformer (KIT) used to implement guidance through body parts. The proposed model has the advantage to effectively overcome the problems brought by the data variances cross multiple species. Our proposed method outperforms the state-of-the-art in animal pose estimation competition at MMVRAC 2024. The source code is public available at https://github.com/Raojiyong/icme_animalpose