Physically Plausible Human-Object Rendering From Sparse Views via 3D Gaussian Splatting.

Area of Science:

• Computer Vision
• Computer Graphics
• Artificial Intelligence

Background:

• Rendering realistic human-object interactions (HOIs) from limited viewpoints is difficult.
• Current methods often fail to balance rendering quality, physical plausibility, and computational speed.

Purpose of the Study:

• To introduce HOGS (Human-Object Rendering via 3D Gaussian Splatting), a novel framework for efficient and physically plausible HOI rendering from sparse views.
• To address the challenges of geometric consistency and interaction realism in HOI rendering.

Main Methods:

• Representing humans and objects as dynamic 3D Gaussians.
• Implementing a novel optimization process directly on 3D Gaussians to enforce geometric consistency and prevent interpenetration.
• Utilizing pre-trained modules for optimization-guided human pose refinement and human-object contact prediction to handle sparse-view ambiguity and guide interaction losses.

Main Results:

• Achieved state-of-the-art rendering quality in human-object and hand-object interaction datasets.
• Demonstrated high computational efficiency compared to existing methods.
• Successfully enforced physical plausibility through novel geometric constraints and optimization techniques.

Conclusions:

• HOGS offers an efficient and effective solution for realistic HOI rendering from sparse views.
• The framework successfully integrates geometric constraints and interaction prediction for physically plausible results.
• HOGS advances the state-of-the-art in rendering complex human-object dynamics.