VIRGi: View-dependent Instant Recoloring of 3D Gaussians Splats.

Area of Science:

• Computer Vision
• Computer Graphics
• Photorealistic Rendering

Background:

• 3D Gaussian Splatting (3DGS) excels at novel view synthesis and 3D reconstruction.
• Existing 3DGS methods lack efficient and photorealistic appearance editing capabilities.

Purpose of the Study:

• Introduce VIRGi, a novel approach for rapid color editing of 3DGS scenes.
• Preserve view-dependent effects like specular highlights during color editing.

Main Methods:

• Develop a novel architecture separating color into diffuse and view-dependent components.
• Employ a multi-view training strategy integrating image patches from multiple viewpoints.
• Implement a rapid recoloring scheme using a single edited image and MLP fine-tuning.

Main Results:

• Achieve seamless color propagation across the entire 3DGS scene in approximately two seconds.
• Demonstrate real-time interaction and control over view-dependent effect strength.
• Show significant quantitative and qualitative advancements over Neural Radiance Fields (NeRF) based methods.

Conclusions:

• VIRGi offers an efficient and photorealistic solution for editing 3DGS scene appearance.
• The method facilitates real-time interaction and precise control over scene recoloring.
• VIRGi represents a significant advancement in 3D scene editing techniques.