Abstract
Reconstructing high-fidelity 3D models of highly articulated animals, such as dogs, from a single in-the-wild image remains a formidable challenge. In this paper, we introduce CORGI, a novel framework for consistency-aware 3D dog reconstruction from a single input image that completely eliminates the need for 3D supervision. To overcome generative inconsistencies and the lack of multi-view capture, our pipeline introduces three core components.
First, we propose a Canonical-Driven Orbital Generation (CDOG) strategy, utilizing specialized Canonical and Orbit LoRAs to normalize arbitrary input poses and synthesize reliable pseudo multi-view images. Second, we design a Consistency-aware Deformable 3DGS (CA-3DGS) module that anchors to D-SMAL, a parametric dog prior, explicitly modeling per-view generative errors through dedicated neural deformation fields to learn accurate vertex-level displacements. Finally, to eliminate structural distortions and recover high-frequency details, we introduce a self-supervised Deformation-Conditioned Generative Repair (DCGR) module.
Extensive experiments demonstrate that CORGI achieves state-of-the-art performance, generalizing seamlessly across diverse dog breeds to produce geometrically accurate, visually coherent, and fully animatable 3D assets ready for downstream applications.
Pipeline
Overview of the CORGI reconstruction pipeline.
Overview of the CORGI framework. From a single input image, CORGI reconstructs a high-fidelity, animatable 3D dog without 3D supervision. (a) CDOG normalizes the input pose and synthesizes reliable 360-degree pseudo-multi-view images. (b) CA-3DGS lifts these 2D observations into a deformable 3DGS field anchored to a D-SMAL template, using neural deformation fields to explicitly isolate view-dependent generative errors. (c) DCGR leverages rendered deformation maps to geometrically condition a diffusion model, rectifying residual artifacts and recovering high-frequency details in a self-supervised manner.
Summary Video
Results
CORGI achieves state-of-the-art performance on single-image 3D dog reconstruction, generalizing across diverse dog breeds, poses, viewpoints, and in-the-wild backgrounds. The reconstructed assets are geometrically accurate, visually coherent, and fully animatable, making them ready for downstream applications.
BibTeX
@article{wu2026corgi,
title={CORGI: Consistency-Aware 3D Dog Reconstruction from a Single Image in the Wild},
author={Wu, Yuxiao and Li, Weile and Zhu, Boyi and Liu, Yumeng and Cai, Youcheng and Liu, Ligang},
journal={arXiv preprint arXiv:2607.00321},
year={2026}
}