MorphGS: Morphology-Adaptive Articulated 3D Motion Transfer from Videos

2026-07-08 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Gaming & Interactive Media · Depth: Expert, extended

Summary

CAMO is a category-agnostic framework designed for transferring articulated 3D motion directly from monocular 2D videos to diverse target meshes. It bypasses traditional reconstruct-then-retarget pipelines and their reliance on category-specific parametric templates like SMPL or SMAL. The core of CAMO involves a morphology-parameterized articulated 3D Gaussian splatting model, combined with dense semantic correspondences, to jointly optimize shape and pose. This approach effectively mitigates shape-pose ambiguities, enabling visually faithful motion transfer across varied object categories, including humanoids, quadrupeds, and non-standard animals. Experimental results show superior motion accuracy and visual coherence, achieving up to 85% reductions in PMD and FID on challenging categories compared to existing methods, and operates efficiently in under 10 minutes on a single RTX 4090 GPU.

Key takeaway

For 3D animators or ML engineers developing character animation pipelines, CAMO offers a robust solution for motion transfer from casual monocular videos, eliminating the need for expensive 3D data or category-specific templates. You can now animate diverse 3D assets, including AI-generated meshes, with high fidelity and efficiency. This approach reduces development complexity and broadens the scope of animatable characters, making advanced 3D animation more accessible.

Key insights

CAMO directly transfers 3D motion from monocular 2D video to diverse meshes using morphology-adaptive Gaussian splatting and semantic correspondences.

Principles

• Direct 2D-to-3D optimization mitigates error propagation from intermediate 3D reconstruction.
• Morphology parameterization disentangles structural variations from pose dynamics.
• Dense semantic correspondences reduce shape-pose ambiguity for coherent pose recovery.

Method

CAMO encapsulates a target mesh with Articulated-GS, parameterizes morphology using learnable bone lengths, global scale, and local Gaussian offsets, then jointly optimizes all parameters via differentiable rendering and dense semantic correspondences.

In practice

• Animate meshes synthesized by generative models like Hunyuan3D.
• Transfer motion between different animal species or broader categories.
• Enable video-to-video transfer for Gaussian-based 3D representations.

Topics

• 3D Motion Transfer
• Gaussian Splatting
• Monocular Video Analysis
• Category-Agnostic Animation
• Morphology Adaptation
• Semantic Correspondence

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• RigPAPR: Rig-Based Animation of Static Neural Point Clouds from a Fixed-Viewpoint Video — Interpolation-based neural point rendering inherently avoids joint-boundary artifacts in rigged animation, unlike fixed-shape primitives.
• Towards Consistent Video Geometry Estimation — ViGeo uses dynamic chunking attention and data refinement for consistent video geometry estimation across various inference modes.
• ParaScale: Scale-Calibrated Camera-Motion Transfer via a Gauge-Invariant Parallax Number — Scale-faithful camera-motion transfer requires preserving the gauge-invariant Parallax Number Pi, not raw trajectories.
• Segment Anything with Motion, Geometry, and Semantic Adaptation for Complex Nonlinear Visual Object Tracking — SAMOSA adapts SAM 2 for robust visual object tracking by integrating explicit motion, geometry, and semantic cues.
• CamDirector: Towards Long-Term Coherent Video Trajectory Editing — CamDirector enhances video trajectory editing with precise camera control and long-term consistency via hybrid warping and history-guided autoregressive diffusion.
• MoVerse: Real-Time Video World Modeling with Panoramic Gaussian Scaffold — MoVerse creates interactive 3D worlds from single images by separating world construction from observation rendering using panoramic diffusion and Gaussian scaffolds.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.