WildAni4D: Towards 4D Animal Mesh Reconstruction

Gyeongsu Cho, Hezhen Hu, Donghyeon Soon, Changwoo Kang, Kyungdon Joo; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Findings, 2026, pp. 160-169

Recovering 4D animal motion, including 3D geometry and global trajectory, is essential for quantitative biomechanics and behavioral analysis. However, two major challenges hinder the progress of animal motion recovery. Existing methods lack sufficient annotated video data and suffer from per-frame temporal instability. To address this, we introduce WildAni4D, a framework that unites a novel synthetic video generation pipeline with a new reconstruction model. First, our generator creates a large-scale dataset of realistic, appearance-consistent video sequences with ground-truth animal geometry and camera trajectories. Second, our reconstruction model robustly estimates temporally coherent motion using a single sequence-level shape and per-frame pose predictions. We demonstrate that our model outperforms state-of-the-art per-frame methods, drastically reducing temporal pose flicker and shape drift. WildAni4D offers a scalable solution for 4D animal reconstruction, enabling large-scale motion analysis from in-the-wild videos. Moreover, WildAni4D enables diverse downstream applications, including animal motion data annotation, animatable animal reconstruction, and text-to-motion generation.