Progressive Pose-Guided 4D Animal Reconstruction from Monocular Video

Siyuan Li1, Weiying Chen1, Yilin Wang1, Xinxin Zuo2, Xingyu Li1, Li Cheng1
1University of Alberta    2Concordia University
ECCV 2026
arXiv Code
Monocular input videos and reconstructed 4D animal outputs
Given monocular videos of animals, our method produces high-fidelity 4D models enabling free-viewpoint rendering across time and viewing angles.

Abstract

Reconstructing 4D animals from monocular videos is challenging due to large inter-species variation, complex articulations, and the lack of reliable templates. Existing approaches typically rely on either strict category-specific priors that restrict generalization, or unconstrained generative models that sacrifice input fidelity. To bridge this gap, we present a progressive test-time optimization framework built on 3D Gaussian Splatting for high-fidelity 4D animal reconstruction from a single video. Our key insight is that a coarse shape prior suffices when coupled with a progressive strategy that disentangles articulated pose from non-rigid deformation. Specifically, we employ a symmetry-aware temporal encoding that exploits bilateral cues while absorbing camera estimation drift and a part-conditioned deformation mechanism guided by learnable part anchors and a learnable skinning field. Extensive experiments demonstrate that our approach generalizes robustly across diverse species, achieving superior geometric accuracy, temporal consistency, and visual fidelity compared to existing baselines, even under severe prior mismatch.

Method Overview

Pipeline for progressive pose-guided 4D animal reconstruction
From a monocular video, we initialize canonical 3D Gaussians and a learnable skinning field from the Fauna prior. Symmetry-aware pose refinement uses learnable part anchors and temporal encoding to estimate per-joint transformations. Part-conditioned deformation aggregates the anchors into per-Gaussian embeddings and queries part-level temporal features to capture fine-grained dynamics for rendering from arbitrary viewpoints and time steps.

Results Gallery

Comparison

Application

Limitations

BibTeX

@misc{li2026progressiveposeguided4danimal,
  title={Progressive Pose-Guided 4D Animal Reconstruction from Monocular Video},
  author={Siyuan Li and Weiying Chen and Yilin Wang and Xinxin Zuo and Xingyu Li and Li Cheng},
  year={2026},
  eprint={2607.00157},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2607.00157}
}