TR-Gaussians: High-Fidelity Real-Time Rendering of Planar Transmission and Reflection With 3D Gaussian Splatting.

Area of Science:

• Computer Vision
• Computer Graphics
• Rendering Techniques

Background:

• Rendering complex visual effects like transmission and reflection in 3D scenes is challenging.
• Existing methods struggle to accurately capture view-dependent appearance, especially with planar elements like glass.

Purpose of the Study:

• To introduce a novel 3D-Gaussian-based representation for high-fidelity rendering of planar transmission and reflection.
• To enable real-time novel view synthesis in indoor scenes with complex optical effects.

Main Methods:

• Proposed Transmission-Reflection Gaussians (TRGaussians), combining 3D Gaussians with learnable reflection planes.
• Modeled transmission with 3D Gaussians and reflections with mirrored Gaussians, blended using a Fresnel-based weighting scheme.
• Developed a multistage optimization framework with color/geometry constraints and opacity perturbation.

Main Results:

• TRGaussians achieve real-time, high-fidelity novel view synthesis.
• The method accurately captures complex appearance effects under varying viewpoints.
• Demonstrated superior performance over state-of-the-art approaches quantitatively and qualitatively.

Conclusions:

• TRGaussians offer a powerful new representation for rendering scenes with planar transmission and reflection.
• The approach significantly advances the state-of-the-art in realistic 3D scene synthesis.
• Enables efficient and accurate rendering of complex visual phenomena previously difficult to achieve.